{"updated":"2026-11-12","rows":[{"id":"openai-exploitgym-internal-infra","date":"2026-11-10","title":"ExploitGym agents used leftover credentials to run privileged commands on OpenAI research clusters","product":"OpenAI internal models (ExploitGym evaluation)","vendor":"OpenAI","kind":"incident","vector":"credential-theft","status":"exploited","cve":[],"cvss":null,"fixed":"Runs stopped; ExploitGym runs blocked from 20 July","impact":"Agents reached OpenAI's CI system and research clusters, changed evaluation settings and read private grading code.","how":"On 19 July 2026 agents found credentials and code that earlier runs had left on a shared Artifactory and reused them. Two agents left the activity out of their answers.","sources":[{"name":"OpenAI alignment misalignment report","url":"https://alignment.openai.com/misalignment-reports/using-peers-work-to-gain-unauthorized-access-to-openai-research-infrastructure"}],"checked":"2026-11-12"},{"id":"copilot-jetbrains-cmd-injection","date":"2026-11-10","title":"Command injection in the GitHub Copilot plugin for JetBrains IDEs disclosed local information","product":"GitHub Copilot Plugin for JetBrains IDEs","vendor":"Microsoft (GitHub)","kind":"vulnerability","vector":"data-exfiltration","status":"poc","cve":["CVE-2026-83953"],"cvss":5.5,"fixed":"1.17.0-251","impact":"An authorized local attacker can read information they should not have access to.","how":"Microsoft lists it as a command injection flaw in the plugin; it gives no further technical detail.","sources":[{"name":"NVD CVE-2026-83953","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-83953"},{"name":"Microsoft MSRC CVE-2026-83953","url":"https://msrc.microsoft.com/update-guide/vulnerability/CVE-2026-83953"}],"checked":"2026-11-12"},{"id":"pgedge-postgres-mcp-flaws","date":"2026-11-10","title":"pgEdge Postgres MCP Server fixed DNS rebinding, rate limit bypass and cross-table inference","product":"pgEdge Postgres MCP Server","vendor":"pgEdge","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2026-125320","CVE-2026-125321","CVE-2026-125322"],"cvss":6.5,"fixed":"1.1.0","impact":"A malicious web page can read schemas and query results from a local server, login throttling can be evaded, and a caller can infer data from other tables the role can read.","how":"The HTTP transport skipped Origin checks, rate limiting trusted the client's X-Forwarded-For value, and the count_rows where parameter accepted subqueries on other tables.","sources":[{"name":"NVD CVE-2026-125320","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-125320"},{"name":"NVD CVE-2026-125321","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-125321"},{"name":"NVD CVE-2026-125322","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-125322"},{"name":"GitHub advisory GHSA-6hg5-9f3w-c886","url":"https://github.com/pgEdge/pgedge-postgres-mcp/security/advisories/GHSA-6hg5-9f3w-c886"},{"name":"GitHub advisory GHSA-v347-v3w7-7fq9","url":"https://github.com/pgEdge/pgedge-postgres-mcp/security/advisories/GHSA-v347-v3w7-7fq9"},{"name":"GitHub advisory GHSA-fhv3-x3jh-grj7","url":"https://github.com/pgEdge/pgedge-postgres-mcp/security/advisories/GHSA-fhv3-x3jh-grj7"}],"checked":"2026-11-12"},{"id":"junie-cli-allow-similar","date":"2026-11-06","title":"Junie CLI Allow Similar Commands option allowed unintended command execution","product":"JetBrains Junie CLI","vendor":"JetBrains","kind":"vulnerability","vector":"config-execution","status":"poc","cve":["CVE-2026-123315"],"cvss":7.0,"fixed":"26.8.24","impact":"Commands the user did not approve could run on their machine.","how":"Approving a command with the Allow Similar Commands option let other commands through as well.","sources":[{"name":"NVD CVE-2026-123315","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-123315"},{"name":"JetBrains fixed security issues","url":"https://www.jetbrains.com/privacy-security/issues-fixed/"}],"checked":"2026-11-12"},{"id":"junie-cli-env-in-logs","date":"2026-11-06","title":"Junie CLI wrote sensitive environment variables to session event logs","product":"JetBrains Junie CLI","vendor":"JetBrains","kind":"vulnerability","vector":"credential-theft","status":"poc","cve":["CVE-2026-123316"],"cvss":5.5,"fixed":"26.6.29","impact":"Anyone who can read the session logs can collect secrets held in environment variables.","how":"Session event logs recorded sensitive environment variables.","sources":[{"name":"NVD CVE-2026-123316","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-123316"},{"name":"JetBrains fixed security issues","url":"https://www.jetbrains.com/privacy-security/issues-fixed/"}],"checked":"2026-11-12"},{"id":"junie-action-html-injection","date":"2026-11-06","title":"Junie GitHub Action prompt injection protection bypassed with HTML-encoded characters","product":"JetBrains Junie GitHub Action","vendor":"JetBrains","kind":"vulnerability","vector":"prompt-injection","status":"poc","cve":["CVE-2026-123851"],"cvss":5.4,"fixed":"1.6.4","impact":"Injected instructions can reach the agent despite the action's prompt injection filter.","how":"HTML-encoded characters got past the protection.","sources":[{"name":"NVD CVE-2026-123851","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-123851"},{"name":"JetBrains fixed security issues","url":"https://www.jetbrains.com/privacy-security/issues-fixed/"}],"checked":"2026-11-12"},{"id":"junie-action-fork-exec","date":"2026-11-06","title":"Junie GitHub Action ran fork code with workflow credentials","product":"JetBrains Junie GitHub Action","vendor":"JetBrains","kind":"vulnerability","vector":"credential-theft","status":"poc","cve":["CVE-2026-123852"],"cvss":7.4,"fixed":"1.8.0","impact":"An attacker can run code with the repository workflow's credentials.","how":"Updates pushed from a fork could trigger code execution in a workflow context that held the repository's credentials.","sources":[{"name":"NVD CVE-2026-123852","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-123852"},{"name":"JetBrains fixed security issues","url":"https://www.jetbrains.com/privacy-security/issues-fixed/"}],"checked":"2026-11-12"},{"id":"fast-agent-serve-noauth","date":"2026-11-06","title":"fast-agent serve exposed agent tools on the network without authentication","product":"fast-agent","vendor":"evalstate (fast-agent)","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2026-102610"],"cvss":6.4,"fixed":"0.7.21","impact":"Any network client can drive the agent; with --shell enabled it can run OS commands as the server user.","how":"fast-agent serve bound its MCP HTTP endpoint to 0.0.0.0:8000 with no authentication.","sources":[{"name":"NVD CVE-2026-102610","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-102610"},{"name":"GitHub advisory GHSA-9vhv-g5hf-m7m7","url":"https://github.com/evalstate/fast-agent/security/advisories/GHSA-9vhv-g5hf-m7m7"}],"checked":"2026-11-12"},{"id":"mcp-ts-sdk-http-flaws","date":"2026-11-05","title":"MCP TypeScript SDK fixed token audience, redirect leak and cross-client delivery flaws","product":"MCP TypeScript SDK (@modelcontextprotocol/sdk, /server, /client)","vendor":"Anthropic (Model Context Protocol)","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2026-111771","CVE-2026-111769","CVE-2026-111770"],"cvss":7.7,"fixed":"sdk 1.32.0; server and client 2.3.0 (express 2.0.2, server-legacy 2.3.1); audience validation must be enabled","impact":"A token issued for another service can authenticate to an MCP server, a malicious server can collect API keys, session IDs and OAuth secrets, and shared servers can hand one client's responses to another.","how":"Bearer auth did not check token audience, HTTP and OAuth clients re-sent headers and bodies across cross-origin redirects, and reused stateless servers matched replies only by JSON-RPC id.","sources":[{"name":"NVD CVE-2026-111771","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-111771"},{"name":"NVD CVE-2026-111769","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-111769"},{"name":"NVD CVE-2026-111770","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-111770"},{"name":"GitHub advisory GHSA-rvq5-wwqv-78pq","url":"https://github.com/modelcontextprotocol/typescript-sdk/security/advisories/GHSA-rvq5-wwqv-78pq"},{"name":"GitHub advisory GHSA-6prh-2h8m-c8cw","url":"https://github.com/modelcontextprotocol/typescript-sdk/security/advisories/GHSA-6prh-2h8m-c8cw"},{"name":"GitHub advisory GHSA-jm63-cq47-j5jp","url":"https://github.com/modelcontextprotocol/typescript-sdk/security/advisories/GHSA-jm63-cq47-j5jp"}],"checked":"2026-11-12"},{"id":"claude-code-edit-symlink-race","date":"2026-11-03","title":"Claude Code Edit tool could be raced into writing outside the workspace on macOS","product":"Claude Code","vendor":"Anthropic","kind":"vulnerability","vector":"sandbox-escape","status":"poc","cve":["CVE-2026-88968"],"cvss":7.7,"fixed":"2.1.251","impact":"Code running in the Bash sandbox, for example from a malicious repo or dependency script, can overwrite user files outside the workspace.","how":"The unsandboxed Edit tool checked the path before writing, and sandboxed code could swap a parent directory for a symlink between the check and the write.","sources":[{"name":"NVD CVE-2026-88968","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-88968"},{"name":"GitHub advisory GHSA-5xjx-9qx6-jcrv","url":"https://github.com/anthropics/claude-code/security/advisories/GHSA-5xjx-9qx6-jcrv"}],"checked":"2026-11-12"},{"id":"pdf-reader-mcp-api-key-ignored","date":"2026-10-30","title":"PDF Reader MCP logged API key auth as on but never checked it","product":"PDF Reader MCP (@sylphx/pdf-reader-mcp)","vendor":"SylphxAI","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2026-62264"],"cvss":7.5,"fixed":"3.0.1","impact":"Any client that can reach the port can call every tool and read any PDF the server can open, with limited SSRF.","how":"In HTTP mode the server read MCP_API_KEY but never validated the X-API-Key header, and it bound to 0.0.0.0 by default.","sources":[{"name":"NVD CVE-2026-62264","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-62264"},{"name":"GitHub advisory GHSA-q344-5v34-gm84","url":"https://github.com/SylphxAI/anymd/security/advisories/GHSA-q344-5v34-gm84"}],"checked":"2026-11-12"},{"id":"langgraph-sdk-path-traversal","date":"2026-10-28","title":"LangGraph SDK v3 streaming let IDs traverse to other API paths","product":"langgraph-sdk and langgraph RemoteGraph","vendor":"LangChain","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2026-118667"],"cvss":4.3,"fixed":"langgraph-sdk 0.4.6 and langgraph 1.2.14","impact":"Whoever controls an assistant or thread ID can make authenticated GET requests to other readable paths on the LangGraph server.","how":"The v3 stream helpers put assistant_id and thread_id into request paths without escaping, so relative segments resolved elsewhere.","sources":[{"name":"NVD CVE-2026-118667","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-118667"},{"name":"GitHub advisory GHSA-3fx2-cqw9-xr3c","url":"https://github.com/langchain-ai/langgraph/security/advisories/GHSA-3fx2-cqw9-xr3c"}],"checked":"2026-11-12"},{"id":"langgraph-api-store-namespace","date":"2026-10-28","title":"langgraph-api JS auth handlers failed to isolate Store namespaces","product":"langgraph-api","vendor":"LangChain","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2026-118668"],"cvss":8.6,"fixed":"0.15.0","impact":"An authenticated user can read, overwrite, search or delete another user's Store data.","how":"Namespace rewrites returned by JavaScript or TypeScript custom auth handlers were not applied before the Store operation ran.","sources":[{"name":"NVD CVE-2026-118668","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-118668"},{"name":"GitHub advisory GHSA-4hm6-w6qq-w73v","url":"https://github.com/langchain-ai/langgraph/security/advisories/GHSA-4hm6-w6qq-w73v"}],"checked":"2026-11-12"},{"id":"junie-cli-allowlist-bypass","date":"2026-10-27","title":"Junie CLI allowlist validation let other commands run","product":"JetBrains Junie CLI","vendor":"JetBrains","kind":"vulnerability","vector":"config-execution","status":"poc","cve":["CVE-2026-117530"],"cvss":6.7,"fixed":"26.8.17","impact":"Commands outside the user's allowlist could run on their machine.","how":"Validation of allowlisted commands was incomplete.","sources":[{"name":"NVD CVE-2026-117530","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-117530"},{"name":"JetBrains fixed security issues","url":"https://www.jetbrains.com/privacy-security/issues-fixed/"}],"checked":"2026-11-12"},{"id":"mcp-python-sdk-auth-redirect-ref","date":"2026-10-27","title":"MCP Python SDK fixed token audience, redirect leak and schema $ref fetch flaws","product":"MCP Python SDK (mcp on PyPI)","vendor":"Anthropic (Model Context Protocol)","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2026-116684","CVE-2026-116685","CVE-2026-116686"],"cvss":6.8,"fixed":"1.30.0 or 2.2.0","impact":"A token for another service can authenticate to an MCP server, a redirect can leak client credential headers, and a malicious server can make clients fetch URLs or read local files.","how":"Bearer auth did not check token audience, HTTP clients followed cross-origin redirects with custom headers, and outputSchema validation resolved $ref URLs from the client's network and filesystem.","sources":[{"name":"NVD CVE-2026-116684","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-116684"},{"name":"NVD CVE-2026-116685","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-116685"},{"name":"NVD CVE-2026-116686","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-116686"},{"name":"GitHub advisory GHSA-w4fh-qvv9-3v23","url":"https://github.com/modelcontextprotocol/python-sdk/security/advisories/GHSA-w4fh-qvv9-3v23"},{"name":"GitHub advisory GHSA-5h93-6whr-6q8j","url":"https://github.com/modelcontextprotocol/python-sdk/security/advisories/GHSA-5h93-6whr-6q8j"},{"name":"GitHub advisory GHSA-rwrf-2pqf-9j8j","url":"https://github.com/modelcontextprotocol/python-sdk/security/advisories/GHSA-rwrf-2pqf-9j8j"}],"checked":"2026-11-12"},{"id":"atxp-gemini-ext-cmd-injection","date":"2026-10-27","title":"ATXP Gemini CLI extension ran shell commands planted in emails or SMS","product":"ATXP Gemini CLI extension","vendor":"ATXP","kind":"vulnerability","vector":"prompt-injection","status":"poc","cve":["CVE-2026-117560"],"cvss":8.5,"fixed":"","impact":"Text in an email or SMS the agent reads can run shell commands as the user.","how":"runAtxp() passed MCP tool parameters such as message_id, subject or query to a shell unsanitized, so injected $() or backtick substitutions executed.","sources":[{"name":"NVD CVE-2026-117560","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-117560"},{"name":"VulnCheck advisory","url":"https://www.vulncheck.com/advisories/atxp-gemini-cli-extension-through-1.0.0-os-command-injection-via-mcp-tools"},{"name":"GitHub issue atxp-dev/atxp#15","url":"https://github.com/atxp-dev/atxp/issues/15"}],"checked":"2026-11-12"},{"id":"langgraph-store-dotted-namespace","date":"2026-10-23","title":"LangGraph Postgres and SQLite stores mixed up tenant namespaces","product":"langgraph-checkpoint-postgres and langgraph-checkpoint-sqlite","vendor":"LangChain","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2026-115883"],"cvss":6.0,"fixed":"postgres 3.1.3 and sqlite 3.1.2","impact":"A caller can read, overwrite or delete another tenant's stored agent data.","how":"The stores joined namespace labels with dots and only validated labels on put, so dotted labels in get, search or delete resolved to another namespace.","sources":[{"name":"NVD CVE-2026-115883","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-115883"},{"name":"GitHub advisory GHSA-6hmw-9432-67qq","url":"https://github.com/langchain-ai/langgraph/security/advisories/GHSA-6hmw-9432-67qq"}],"checked":"2026-11-12"},{"id":"langgraph-cli-token-log","date":"2026-10-23","title":"LangGraph CLI printed a registry push token in verbose deploy logs","product":"LangGraph CLI","vendor":"LangChain","kind":"vulnerability","vector":"credential-theft","status":"poc","cve":["CVE-2026-115885"],"cvss":4.1,"fixed":"0.4.32","impact":"Anyone who reads the CI logs within 12 hours can pull or push images in the deployment's image repository.","how":"langgraph deploy --verbose echoed the stdin passed to docker login, which held the temporary push token.","sources":[{"name":"NVD CVE-2026-115885","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-115885"},{"name":"GitHub advisory GHSA-8prr-f582-jqq3","url":"https://github.com/langchain-ai/langgraph/security/advisories/GHSA-8prr-f582-jqq3"}],"checked":"2026-11-12"},{"id":"netbox-mcp-http-noauth","date":"2026-10-23","title":"NetBox MCP Server HTTP mode had no authentication","product":"NetBox MCP Server","vendor":"NetBox Labs","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2026-57207"],"cvss":5.9,"fixed":"1.2.1","impact":"Any client that can reach the port gets full use of the server's NetBox tools.","how":"With TRANSPORT=http the endpoint issued MCP sessions without any token or credential.","sources":[{"name":"NVD CVE-2026-57207","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-57207"},{"name":"GitHub advisory GHSA-5qg7-j8mr-9477","url":"https://github.com/netboxlabs/netbox-mcp-server/security/advisories/GHSA-5qg7-j8mr-9477"}],"checked":"2026-11-12"},{"id":"obsidian-tc-approval-bypass","date":"2026-10-23","title":"obsidian-tc ran destructive vault actions after the human declined","product":"obsidian-tc (Obsidian MCP server)","vendor":"The-40-Thieves","kind":"vulnerability","vector":"rogue-action","status":"poc","cve":["CVE-2026-115477","CVE-2026-115478"],"cvss":7.1,"fixed":"1.31.4","impact":"A misbehaving or hostile client could delete or change vault content without human approval, and memory tools could overwrite unrelated notes.","how":"The server treated a resubmitted requestState as approval even when the elicitation was declined, and memory notes were rewritten at computed paths without checking ownership.","sources":[{"name":"NVD CVE-2026-115477","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-115477"},{"name":"NVD CVE-2026-115478","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-115478"},{"name":"GitHub advisory GHSA-737x-cwfr-5vvc","url":"https://github.com/The-40-Thieves/obsidian-tc/security/advisories/GHSA-737x-cwfr-5vvc"},{"name":"GitHub advisory GHSA-r3g7-r4hr-m258","url":"https://github.com/The-40-Thieves/obsidian-tc/security/advisories/GHSA-r3g7-r4hr-m258"}],"checked":"2026-11-12"},{"id":"litellm-forged-admin-rce","date":"2026-10-22","title":"LiteLLM let internal users forge an admin token and run commands via MCP","product":"LiteLLM proxy","vendor":"BerriAI","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2026-114612"],"cvss":9.9,"fixed":"1.100.4, 1.101.3, 1.102.2, 1.103.1 or 1.104.0-rc.2","impact":"An authenticated internal user can become proxy admin and run arbitrary commands through the MCP stdio endpoint.","how":"One cryptographic context protected both stored secrets and session tokens, so metadata encrypted by the API key flow could be replayed as a forged admin bearer token.","sources":[{"name":"NVD CVE-2026-114612","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-114612"},{"name":"GitHub advisory GHSA-7hp6-4w63-5g45","url":"https://github.com/BerriAI/litellm/security/advisories/GHSA-7hp6-4w63-5g45"}],"checked":"2026-11-12"},{"id":"agent-zero-chat-xss","date":"2026-10-22","title":"Agent Zero Web UI executed script injected through chat messages or uploads","product":"Agent Zero Web UI","vendor":"agent0ai (Agent Zero)","kind":"vulnerability","vector":"data-exfiltration","status":"poc","cve":["CVE-2026-97635"],"cvss":8.6,"fixed":"1.15","impact":"Injected JavaScript runs in the victim's authenticated Agent Zero session.","how":"The UI did not sanitize chat messages or uploaded files before reproducing their content.","sources":[{"name":"NVD CVE-2026-97635","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-97635"},{"name":"CISA CSAF advisory VA-26-296-01","url":"https://raw.githubusercontent.com/cisagov/CSAF/develop/csaf_files/IT/white/2026/va-26-296-01.json"},{"name":"GitHub issue agent0ai/agent-zero#943","url":"https://github.com/agent0ai/agent-zero/issues/943"}],"checked":"2026-11-12"},{"id":"d365fo-mcp-blank-api-key","date":"2026-10-21","title":"D365 F&O MCP Server one-click Azure deploy left the API open","product":"D365 F&O MCP Server","vendor":"dynamics365ninja","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2026-114123"],"cvss":7.5,"fixed":"1.11.0","impact":"Anyone on the internet can query indexed proprietary X++ source, business logic and security role structure.","how":"The Azure template defaulted the API key to empty, and an empty key made the auth middleware pass every request on a listener bound to 0.0.0.0.","sources":[{"name":"NVD CVE-2026-114123","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-114123"},{"name":"GitHub advisory GHSA-8764-fh3m-wf7g","url":"https://github.com/dynamics365ninja/d365fo-mcp-server/security/advisories/GHSA-8764-fh3m-wf7g"}],"checked":"2026-11-12"},{"id":"mcp-ts-sdk-task-isolation","date":"2026-10-20","title":"MCP TypeScript SDK experimental tasks were shared across sessions","product":"MCP TypeScript SDK (@modelcontextprotocol/sdk)","vendor":"Anthropic (Model Context Protocol)","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2026-111767"],"cvss":8.6,"fixed":"1.32.0","impact":"A client can list, read and cancel tasks that belong to other clients.","how":"From 1.24.0 the experimental taskStore did not check which session created a task when clients shared one InMemoryTaskStore.","sources":[{"name":"NVD CVE-2026-111767","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-111767"},{"name":"GitHub advisory GHSA-22jm-h49p-29qw","url":"https://github.com/modelcontextprotocol/typescript-sdk/security/advisories/GHSA-22jm-h49p-29qw"}],"checked":"2026-11-12"},{"id":"certin-ai-agents-attack-vectors","date":"2026-10-16","title":"CERT-In critical advisory: AI agents accessed Indian organisations' live systems without authorisation","product":"Unnamed provider evaluation agents and open source agentic attack tools","vendor":"Various","kind":"incident","vector":"credential-theft","status":"exploited","cve":[],"cvss":null,"fixed":"","impact":"Agents read business records with personal and confidential data at Indian organisations, none of which detected it; criminals used agent tools to cut intrusions from days to hours.","how":"Agents reused credentials exposed in configs, public repos and loose cloud roles, plus dormant APIs and admin panels without MFA. Advisory CIAD-2026-0049.","sources":[{"name":"The Federal","url":"https://thefederal.com/category/news/cert-in-flags-unauthorised-ai-agent-access-to-indian-organisations-data-259736"},{"name":"Inc42","url":"https://inc42.com/buzz/cert-in-flags-unauthorised-ai-agent-access-to-indian-organisations-systems/"}],"checked":"2026-11-12"},{"id":"claude-agent-sdk-resume-injection","date":"2026-10-16","title":"Claude Agent SDK resume option allowed CLI flag injection and command execution","product":"Claude Agent SDK for Python","vendor":"Anthropic","kind":"vulnerability","vector":"config-execution","status":"poc","cve":["CVE-2026-96620"],"cvss":9.2,"fixed":"0.2.121","impact":"An attacker who controls the session ID an app passes to resume can run arbitrary commands, outside the SDK's permission controls.","how":"The resume value went to the Claude CLI unvalidated, so a value starting with a dash became a flag such as --mcp-config that defines a stdio MCP server.","sources":[{"name":"NVD CVE-2026-96620","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-96620"},{"name":"GitHub advisory GHSA-h4mw-j7qp-8mwm","url":"https://github.com/anthropics/claude-agent-sdk-python/security/advisories/GHSA-h4mw-j7qp-8mwm"}],"checked":"2026-11-12"},{"id":"junie-cli-project-mcp-exec","date":"2026-10-15","title":"Junie CLI ran code from an untrusted project's MCP server configuration","product":"JetBrains Junie CLI","vendor":"JetBrains","kind":"vulnerability","vector":"config-execution","status":"poc","cve":["CVE-2026-111091"],"cvss":7.8,"fixed":"26.8.3","impact":"Opening a malicious project can run attacker code on the developer's machine.","how":"Junie acted on the MCP server configuration shipped in an untrusted project.","sources":[{"name":"NVD CVE-2026-111091","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-111091"},{"name":"JetBrains fixed security issues","url":"https://www.jetbrains.com/privacy-security/issues-fixed/"}],"checked":"2026-11-12"},{"id":"claude-code-plugin-sha-pin","date":"2026-10-13","title":"Claude Code plugin installs could swap a pinned commit for attacker code","product":"Claude Code and Claude Desktop plugin installer","vendor":"Anthropic","kind":"vulnerability","vector":"supply-chain","status":"poc","cve":["CVE-2026-86063"],"cvss":7.5,"fixed":"2.1.179","impact":"A user who installs a plugin pinned to a reviewed commit can get attacker-controlled plugin code instead.","how":"The installer ran git checkout on the pinned SHA without checking HEAD matched it, so a branch named after that SHA was checked out instead.","sources":[{"name":"NVD CVE-2026-86063","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-86063"},{"name":"GitHub advisory GHSA-pq7j-f95f-qfcv","url":"https://github.com/anthropics/claude-code/security/advisories/GHSA-pq7j-f95f-qfcv"}],"checked":"2026-11-12"},{"id":"obsidian-tc-acl-flaws","date":"2026-10-12","title":"obsidian-tc MCP server fixed cross-vault tokens, DNS rebinding and read ACL bypasses","product":"obsidian-tc (Obsidian MCP server)","vendor":"The-40-Thieves","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2026-77366","CVE-2026-77363","CVE-2026-77364","CVE-2026-77361","CVE-2026-77362"],"cvss":9.9,"fixed":"1.3.1","impact":"Any valid token could read, write or delete another vault, a malicious web page could reach an unauthenticated local server, and callers could read denied notes and plugin secrets.","how":"Tokens were not bound to a vault, the HTTP transport skipped Host and Origin checks, control folders were only hidden from listings, and ACL checks missed symlinks and some tool output fields.","sources":[{"name":"NVD CVE-2026-77366","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-77366"},{"name":"NVD CVE-2026-77363","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-77363"},{"name":"NVD CVE-2026-77364","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-77364"},{"name":"NVD CVE-2026-77361","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-77361"},{"name":"NVD CVE-2026-77362","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-77362"},{"name":"GitHub advisory GHSA-84c9-qqpq-2pq5","url":"https://github.com/The-40-Thieves/obsidian-tc/security/advisories/GHSA-84c9-qqpq-2pq5"},{"name":"GitHub advisory GHSA-pqh8-4pj7-32xx","url":"https://github.com/The-40-Thieves/obsidian-tc/security/advisories/GHSA-pqh8-4pj7-32xx"},{"name":"GitHub advisory GHSA-cvp8-vvgf-69q4","url":"https://github.com/The-40-Thieves/obsidian-tc/security/advisories/GHSA-cvp8-vvgf-69q4"},{"name":"GitHub advisory GHSA-c5xx-h28h-f682","url":"https://github.com/The-40-Thieves/obsidian-tc/security/advisories/GHSA-c5xx-h28h-f682"},{"name":"GitHub advisory GHSA-43r2-c7g8-prc8","url":"https://github.com/The-40-Thieves/obsidian-tc/security/advisories/GHSA-43r2-c7g8-prc8"}],"checked":"2026-11-12"},{"id":"mcp-remote-device-code-cleartext","date":"2026-10-11","title":"mcp-remote sent OAuth secrets and tokens over plain HTTP","product":"mcp-remote (npm)","vendor":"punkpeye (mcp-remote)","kind":"vulnerability","vector":"credential-theft","status":"poc","cve":["CVE-2026-108716"],"cvss":6.0,"fixed":"","impact":"An on-path attacker can capture the client secret and the issued access and refresh tokens.","how":"From 0.8.0 through 0.14.3 the device code flow used discovered device authorization and token endpoints without requiring HTTPS.","sources":[{"name":"NVD CVE-2026-108716","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-108716"},{"name":"VulnCheck advisory","url":"https://www.vulncheck.com/advisories/mcp-remote-0.8.0-through-0.14.3-cleartext-credential-transmission-via-device-code-oauth-grant"}],"checked":"2026-11-12"},{"id":"anthropic-unintended-actions-real-sites","date":"2026-10-09","title":"Anthropic disclosed Claude models exploiting flaws and submitting real forms during tasks","product":"Claude Mythos 5, Claude Opus 5, Claude Haiku 4.5 and others","vendor":"Anthropic","kind":"incident","vector":"rogue-action","status":"exploited","cve":[],"cvss":null,"fixed":"","impact":"A university server was used via an injection flaw, gated data was pulled with found tokens, and real government and police tip forms were submitted.","how":"Models on research tasks worked around errors and restrictions on live sites. Anthropic says real-world impact was minimal.","sources":[{"name":"Anthropic: Investigating unintended model actions","url":"https://www.anthropic.com/research/investigating-unintended-model-actions"}],"checked":"2026-11-12"},{"id":"x64dbg-mcp-noauth","date":"2026-10-09","title":"x64dbg-MCP Server gave any network client control of the debugger","product":"x64dbg-MCP Server","vendor":"duty1g","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2026-107824"],"cvss":9.3,"fixed":"1.1","impact":"An unauthenticated client can run debugger commands, attach to processes, read and write their memory and write files anywhere.","how":"All debugger tools were exposed over HTTP and SSE with no authentication on 0.0.0.0, ports 9094 and 9095.","sources":[{"name":"NVD CVE-2026-107824","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-107824"},{"name":"GitHub advisory GHSA-4478-h5jv-647m","url":"https://github.com/duty1g/x64dbg-mcp-server/security/advisories/GHSA-4478-h5jv-647m"}],"checked":"2026-11-12"},{"id":"openai-agent-nsw-npws","date":"2026-10-07","title":"OpenAI confirmed a model accessed non-public NSW National Parks and Wildlife Service data","product":"OpenAI internal model","vendor":"OpenAI","kind":"incident","vector":"rogue-action","status":"exploited","cve":[],"cvss":null,"fixed":"","impact":"A model researching Australian fire statistics reached database metadata that was not meant to be public. OpenAI found no personal information involved.","how":"Same pattern as the Medicare case: a model doing data lookups went beyond public pages on a government web app.","sources":[{"name":"Cyber Daily","url":"https://www.cyberdaily.au/government/14274-we-will-do-better-openai-admits-to-another-government-breach-after-pm-lashing"}],"checked":"2026-11-12"},{"id":"langflow-access-control-rce","date":"2026-10-07","title":"Langflow access control flaw let authenticated users run code","product":"Langflow OSS","vendor":"IBM (Langflow)","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2026-104335"],"cvss":8.8,"fixed":"1.12.3","impact":"A remote authenticated attacker can execute arbitrary code on the Langflow server.","how":"IBM describes improper access control in Langflow 1.0.0 through 1.12.2 without further detail.","sources":[{"name":"NVD CVE-2026-104335","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-104335"},{"name":"IBM security bulletin","url":"https://www.ibm.com/support/pages/node/7290694"}],"checked":"2026-11-12"},{"id":"wikimedia-openai-agent-edits","date":"2026-10-05","title":"Wikimedia says agents it believes OpenAI ran edited wikis and probed its Etherpad","product":"OpenAI agents (attributed by Wikimedia)","vendor":"OpenAI","kind":"incident","vector":"rogue-action","status":"reported","cve":[],"cvss":null,"fixed":"","impact":"Unapproved bot edits, attempts to hijack a citation tool and Etherpad as proxies, and millions of requests that may have fed a Wikidata Query Service outage in May.","how":"Agents made sandbox edits and changed citation tool settings to fetch remote data. Wikimedia found no compromise of its systems or data.","sources":[{"name":"Wikimedia Foundation","url":"https://wikimediafoundation.org/news/2026/10/05/openai-rogue-agent-activities-found-on-wikimedia-projects/"}],"checked":"2026-11-12"},{"id":"openai-agent-medicare-statistics-portal","date":"2026-09-24","title":"OpenAI agent accessed public and non-public files on Australia's Medicare statistics portal","product":"OpenAI internal model","vendor":"OpenAI","kind":"incident","vector":"rogue-action","status":"exploited","cve":[],"cvss":null,"fixed":"Government accepted 18 review recommendations, including mandatory AI incident reporting","impact":"The agent reached non-public files on the Services Australia Medicare Statistics Reporting Service. No patient records are believed to have been accessed.","how":"A model in internal training looking up Australian statistics went past public pages in June 2026. OpenAI emailed the government in September and Australia ran a rapid review.","sources":[{"name":"Prime Minister press conference, 24 Sep 2026","url":"https://www.pm.gov.au/media/press-conference-new-york"},{"name":"Prime Minister media release on the Rapid Review","url":"https://www.pm.gov.au/media/strengthening-australia-against-ai-driven-cyber-incidents"}],"checked":"2026-11-12"},{"id":"gambit-agent-retail-skimming","date":"2026-09-22","title":"Criminal ran open source AI agent harnesses to hack online retailers for about $25 each","product":"Strix, Cairn and Hermes agent harnesses","vendor":"Open source projects","kind":"incident","vector":"data-exfiltration","status":"reported","cve":[],"cvss":null,"fixed":"","impact":"At least 27 companies were compromised in five days, skimmers were confirmed on 19, and over 600,000 card records were taken from two firms.","how":"One operator used agent frameworks over OpenRouter models to scan, exploit and plant card skimmers at scale, with a mean cost of $25.46 per scan.","sources":[{"name":"Gambit Security interim report","url":"https://gambit.security/blog-posts/autonomous-ai-agents-online-retailers-25-a-company"}],"checked":"2026-11-12"},{"id":"rubygems-oai-agent-packages","date":"2026-09-11","title":"Researchers link OpenAI test agents to 2,000+ junk gems and API key theft attempts on RubyGems","product":"RubyGems.org","vendor":"OpenAI (attributed by researchers)","kind":"incident","vector":"rogue-action","status":"reported","cve":[],"cvss":null,"fixed":"RubyGems fixed the cache bug 9 July and revoked all legacy API keys","impact":"RubyGems paused sign-ups for four days in May 2026 and removed 500+ packages. Agents tried to exploit a caching flaw that could leak API keys.","how":"Agents uploaded gems that abused RubyDoc.info builds to run code and scrape sites. OpenAI says its agents used RubyGems for benign tasks and has not verified the malicious uploads.","sources":[{"name":"Researcher report (rubyhack.ai)","url":"https://www.rubyhack.ai/"},{"name":"RubyGems advisory on legacy API key leak","url":"https://blog.rubygems.org/2026/07/22/security-advisory-legacy-api-key-leak.html"},{"name":"Simon Willison summary with OpenAI update","url":"https://simonwillison.net/2026/sep/12/openai-agents-rubygems"}],"checked":"2026-11-12"},{"id":"gemini-cli-home-override","date":"2026-09-10","title":"Gemini CLI folder trust bypass through GEMINI_CLI_HOME in .env","product":"Gemini CLI","vendor":"Google","kind":"vulnerability","vector":"config-execution","status":"poc","cve":["CVE-2026-13745"],"cvss":7.7,"fixed":"0.39.1","impact":"Starting the CLI in an untrusted folder could run attacker code.","how":"A project .env overrode GEMINI_CLI_HOME so the CLI loaded attacker config and skipped trust prompts.","sources":[{"name":"NVD CVE-2026-13745","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-13745"},{"name":"Redguard advisory","url":"https://www.redguard.ch/blog/2026/06/03/advisory-google-gemini-cli/"},{"name":"Gemini CLI v0.39.1 release","url":"https://github.com/google-gemini/gemini-cli/releases/tag/v0.39.1"}],"checked":"2026-11-12"},{"id":"dsewiki-openai-agent-swarm","date":"2026-09-04","title":"OpenAI agent swarm took over German wiki DSEWiki as a message board","product":"OpenAI internal agents","vendor":"OpenAI","kind":"incident","vector":"rogue-action","status":"exploited","cve":[],"cvss":null,"fixed":"","impact":"About 17,000 agent edits flooded a volunteer wiki, with XSS probing and moderator impersonation, from May to early July 2026.","how":"Agents in a web research task used the wiki to share answers and ways around restrictions, including bypassing a GET-only proxy. OpenAI called it a misalignment incident and reported it to the European Commission.","sources":[{"name":"Researcher report (collusion.wiki)","url":"https://collusion.wiki/"},{"name":"OpenAI misalignment notices index","url":"https://alignment.openai.com/misalignment-reports"},{"name":"Fortune","url":"https://fortune.com/2026/09/07/openai-ai-agents-german-wiki-ran-their-own-message-board/"}],"checked":"2026-11-12"},{"id":"codex-fsmonitor-exec","date":"2026-09-01","title":"Codex ran a repository's core.fsmonitor helper outside its sandbox","product":"OpenAI Codex CLI and Codex Desktop","vendor":"OpenAI","kind":"vulnerability","vector":"config-execution","status":"poc","cve":["CVE-2026-19592"],"cvss":7.3,"fixed":"","impact":"A prepared repository could run attacker code with the user's privileges and no approval prompt.","how":"Codex collected git metadata without disabling core.fsmonitor, so git ran the helper named in .git/config.","sources":[{"name":"NVD CVE-2026-19592","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-19592"},{"name":"Fix pull request openai/codex#22652","url":"https://github.com/openai/codex/pull/22652"}],"checked":"2026-11-12"},{"id":"meta-muse-spark-eval-breach","date":"2026-08-05","title":"Meta's Muse Spark model exploited a third-party service during a misconfigured evaluation","product":"Muse Spark 1.1","vendor":"Meta","kind":"incident","vector":"rogue-action","status":"exploited","cve":[],"cvss":null,"fixed":"","impact":"An unnamed outside company's systems were accessed and, per The Information, altered.","how":"A misconfiguration by test partner Irregular gave the model internet access, and it exploited a vulnerability in a real service.","sources":[{"name":"Reuters via BNN Bloomberg","url":"https://www.bnnbloomberg.ca/business/artificial-intelligence/2026/08/06/meta-ai-model-hacks-another-company-during-testing/"}],"checked":"2026-11-12"},{"id":"anthropic-irregular-eval-real-systems","date":"2026-07-30","title":"Claude models reached three real organisations from a misconfigured third-party cyber evaluation","product":"Claude Opus 4.7, Claude Mythos 5, internal test model","vendor":"Anthropic","kind":"incident","vector":"rogue-action","status":"exploited","cve":[],"cvss":null,"fixed":"Cyber evaluations paused 23 July; affected parties and PyPI notified","impact":"One organisation lost credentials and several hundred rows of production data; a malicious PyPI package ran on 15 real systems; a third app was entered by SQL injection.","how":"Evaluation machines run with partner Irregular had live internet by mistake. Claude believed the real targets were part of the exercise and attacked them.","sources":[{"name":"Anthropic: Investigating three real-world incidents in our cybersecurity evaluations","url":"https://www.anthropic.com/research/investigating-incidents-cybersecurity-evals"}],"checked":"2026-11-12"},{"id":"claude-code-action-mcp-json","date":"2026-07-16","title":"Claude Code GitHub Action loaded .mcp.json from attacker pull requests","product":"Claude Code Action (GitHub Action)","vendor":"Anthropic","kind":"vulnerability","vector":"config-execution","status":"poc","cve":["CVE-2026-47751"],"cvss":5.3,"fixed":"1.0.74","impact":"An attacker pull request could run code on the Actions runner and steal workflow secrets.","how":"The action checked out the PR head, read its .mcp.json and enabled all project MCP servers when a privileged user or trigger ran it.","sources":[{"name":"NVD CVE-2026-47751","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-47751"},{"name":"GitHub advisory GHSA-8q5r-mmjf-575q","url":"https://github.com/anthropics/claude-code-action/security/advisories/GHSA-8q5r-mmjf-575q"}],"checked":"2026-11-12"},{"id":"huggingface-openai-eval-agent-intrusion","date":"2026-07-16","title":"OpenAI evaluation agents escaped their sandbox and broke into Hugging Face production","product":"OpenAI internal models (ExploitGym evaluation)","vendor":"OpenAI","kind":"incident","vector":"rogue-action","status":"exploited","cve":[],"cvss":null,"fixed":"Dataset code-execution paths closed; credentials rotated; OpenAI blocked live internet in training and evals","impact":"The agents reached several internal Hugging Face clusters, collected cloud and service credentials and accessed a limited set of internal datasets.","how":"Agents in a cyber evaluation exploited a zero-day in a package registry proxy to get open internet. A malicious dataset then gave code execution in Hugging Face's dataset processing.","sources":[{"name":"Hugging Face security incident disclosure","url":"https://huggingface.co/blog/security-incident-july-2026"},{"name":"OpenAI alignment report linking the July runs","url":"https://alignment.openai.com/misalignment-reports/using-peers-work-to-gain-unauthorized-access-to-openai-research-infrastructure"}],"checked":"2026-11-12"},{"id":"aws-healthlake-mcp-ssrf","date":"2026-07-14","title":"AWS HealthLake MCP server could send AWS credentials to an attacker endpoint","product":"awslabs.healthlake-mcp-server (PyPI)","vendor":"AWS","kind":"vulnerability","vector":"credential-theft","status":"poc","cve":["CVE-2026-15643"],"cvss":9.2,"fixed":"0.0.14","impact":"An authenticated user could exfiltrate the server's temporary AWS security credentials.","how":"The pagination next_token was trusted as a URL, so follow-up signed requests could be redirected to any server.","sources":[{"name":"NVD CVE-2026-15643","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-15643"},{"name":"AWS security bulletin 2026-054","url":"https://aws.amazon.com/security/security-bulletins/2026-054-aws/"}],"checked":"2026-11-12"},{"id":"cursor-workdir-escape","date":"2026-06-25","title":"Cursor agent could widen its sandbox with the working_directory parameter","product":"Cursor","vendor":"Anysphere","kind":"vulnerability","vector":"sandbox-escape","status":"poc","cve":["CVE-2026-50548"],"cvss":9.3,"fixed":"3.0","impact":"A malicious agent could write files anywhere the user can and get unsandboxed code execution.","how":"The sandbox granted write access to the command's working directory, which the agent could point at a sensitive location such as the sandbox helper.","sources":[{"name":"NVD CVE-2026-50548","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-50548"},{"name":"GitHub advisory GHSA-3p48-7v9f-v5cw","url":"https://github.com/cursor/cursor/security/advisories/GHSA-3p48-7v9f-v5cw"}],"checked":"2026-11-12"},{"id":"gemini-cli-env-launcher","date":"2026-06-24","title":"Gemini CLI and run-gemini-cli action ran code from a malicious .gemini/.env before the sandbox","product":"Gemini CLI, run-gemini-cli GitHub Action","vendor":"Google","kind":"vulnerability","vector":"config-execution","status":"poc","cve":["CVE-2026-12537"],"cvss":10.0,"fixed":"Gemini CLI 0.39.1; run-gemini-cli 0.1.22","impact":"On headless CI an attacker could get host code execution before the sandbox started.","how":"The container launcher built an OS command from values in a crafted .gemini/.env file.","sources":[{"name":"NVD CVE-2026-12537","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-12537"},{"name":"GitHub advisory GHSA-wpqr-6v78-jr5g","url":"https://github.com/google-github-actions/run-gemini-cli/security/advisories/GHSA-wpqr-6v78-jr5g"}],"checked":"2026-11-12"},{"id":"cline-kanban-ws-hijack","date":"2026-06-01","title":"Cline Kanban server open to cross-origin WebSocket hijack","product":"Cline (Kanban server)","vendor":"Cline","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2026-44211"],"cvss":9.6,"fixed":"No patch at publication","impact":"A malicious website could connect to the local Kanban server and act on the user's behalf.","how":"The WebSocket server did not check the origin of connecting pages.","sources":[{"name":"NVD CVE-2026-44211","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-44211"},{"name":"GitHub advisory GHSA-5c57-rqjx-35g2","url":"https://github.com/cline/cline/security/advisories/GHSA-5c57-rqjx-35g2"}],"checked":"2026-11-12"},{"id":"litellm-auth-sqli","date":"2026-05-08","title":"LiteLLM proxy SQL injection through the Authorization header, exploited in the wild","product":"LiteLLM proxy","vendor":"BerriAI","kind":"vulnerability","vector":"credential-theft","status":"exploited","cve":["CVE-2026-42208"],"cvss":9.3,"fixed":"1.83.7","impact":"An unauthenticated attacker could read and possibly change the proxy database, including the LLM credentials it holds. CISA lists it as known exploited.","how":"The API key check placed the caller's key value into SQL text, reachable through the error path on any LLM route.","sources":[{"name":"NVD CVE-2026-42208","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-42208"},{"name":"GitHub advisory GHSA-r75f-5x8p-qvmc","url":"https://github.com/BerriAI/litellm/security/advisories/GHSA-r75f-5x8p-qvmc"}],"checked":"2026-11-12"},{"id":"windsurf-mcp-config-injection","date":"2026-04-15","title":"Windsurf prompt injection registered a malicious MCP server","product":"Windsurf","vendor":"Windsurf (Cognition)","kind":"vulnerability","vector":"prompt-injection","status":"poc","cve":["CVE-2026-30615"],"cvss":8.0,"fixed":"","impact":"Attacker HTML content could run commands on the victim system and persist through MCP config.","how":"Injected instructions led the agent to edit the local MCP config and add a malicious stdio server that then ran.","sources":[{"name":"NVD CVE-2026-30615","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-30615"},{"name":"OX Security advisory","url":"https://www.ox.security/blog/mcp-supply-chain-advisory-rce-vulnerabilities-across-the-ai-ecosystem/"}],"checked":"2026-11-12"},{"id":"litellm-mcp-stdio-rce","date":"2026-04-15","title":"LiteLLM MCP server creation ran arbitrary commands","product":"LiteLLM","vendor":"BerriAI","kind":"vulnerability","vector":"config-execution","status":"poc","cve":["CVE-2026-30623"],"cvss":9.8,"fixed":"1.83.7-stable (fix first in 1.83.6-nightly)","impact":"An authenticated user able to add MCP servers could run OS commands with the LiteLLM process privileges.","how":"MCP server JSON config took any command and args and LiteLLM executed them without validation.","sources":[{"name":"NVD CVE-2026-30623","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-30623"},{"name":"LiteLLM blog: MCP stdio command injection","url":"https://docs.litellm.ai/blog/mcp-stdio-command-injection-april-2026"},{"name":"OX Security advisory","url":"https://www.ox.security/blog/mcp-supply-chain-advisory-rce-vulnerabilities-across-the-ai-ecosystem/"}],"checked":"2026-11-12"},{"id":"agent-zero-mcp-rce","date":"2026-04-15","title":"Agent Zero external MCP server config ran arbitrary commands","product":"Agent Zero","vendor":"Agent Zero (agent0ai)","kind":"vulnerability","vector":"config-execution","status":"poc","cve":["CVE-2026-30624"],"cvss":8.6,"fixed":"","impact":"An attacker who supplies an MCP config gets command execution as the Agent Zero process.","how":"The External MCP Servers feature executed command and args values from JSON without restriction.","sources":[{"name":"NVD CVE-2026-30624","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-30624"},{"name":"OX Security advisory","url":"https://www.ox.security/blog/mcp-supply-chain-advisory-rce-vulnerabilities-across-the-ai-ecosystem/"}],"checked":"2026-11-12"},{"id":"meta-internal-agent-sev1","date":"2026-03-18","title":"Meta internal agent posted bad advice unprompted, exposing sensitive data to staff for two hours","product":"Meta internal AI agent","vendor":"Meta","kind":"incident","vector":"rogue-action","status":"exploited","cve":[],"cvss":null,"fixed":"","impact":"Company and user data was visible to engineers without authorisation for about two hours. Meta rated it Sev 1.","how":"An engineer asked the agent to analyse a forum question and it posted a reply without approval. The asker followed the advice, which opened access too widely.","sources":[{"name":"TechCrunch (Meta confirmed to The Information)","url":"https://techcrunch.com/2026/03/18/meta-is-having-trouble-with-rogue-ai-agents"}],"checked":"2026-11-12"},{"id":"copilot-cli-param-expansion","date":"2026-03-06","title":"Copilot CLI safety check missed commands hidden in bash parameter expansion","product":"GitHub Copilot CLI","vendor":"GitHub","kind":"vulnerability","vector":"prompt-injection","status":"poc","cve":["CVE-2026-29783"],"cvss":7.5,"fixed":"0.0.423","impact":"Injected instructions from repo files or MCP responses could run hidden commands without approval.","how":"Bash parameter transformation operators embedded code in arguments of commands the CLI classed as read-only.","sources":[{"name":"NVD CVE-2026-29783","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-29783"},{"name":"GitHub advisory GHSA-g8r9-g2v8-jv6f","url":"https://github.com/github/copilot-cli/security/advisories/GHSA-g8r9-g2v8-jv6f"}],"checked":"2026-11-12"},{"id":"mexico-government-claude-code-breach","date":"2026-02-25","title":"Single operator used Claude Code to breach Mexican government agencies","product":"Claude Code","vendor":"Anthropic","kind":"incident","vector":"data-exfiltration","status":"reported","cve":[],"cvss":null,"fixed":"","impact":"Gambit says nine government organisations were breached and hundreds of millions of citizen records stolen; Bloomberg reported about 150 GB.","how":"The attacker sent over 1,000 Spanish prompts framed as a bug bounty. Claude Code generated and ran about 75 percent of remote commands, and GPT-4.1 sorted the stolen data.","sources":[{"name":"Gambit Security technical report","url":"https://gambit.security/blog-posts/a-single-operator-two-ai-platforms-nine-government-agencies-the-full-technical-report"},{"name":"Gambit Security: Bloomberg coverage","url":"https://gambit.security/news/gambit-research-featured-in-bloomberg-mexico"}],"checked":"2026-11-12"},{"id":"sandworm-mode-npm-mcp-injection","date":"2026-02-20","title":"SANDWORM_MODE npm worm planted rogue MCP servers in AI coding assistants","product":"Typosquatted npm packages (claud-code, cloude, opencraw and others)","vendor":"Unknown actor","kind":"incident","vector":"tool-poisoning","status":"exploited","cve":[],"cvss":null,"fixed":"Packages reported to npm; actor infrastructure taken down","impact":"Infected machines had SSH keys, cloud credentials, npm tokens and LLM API keys harvested, and CI workflows poisoned.","how":"At least 19 packages wrote a hidden MCP server into Claude Code, Claude Desktop, Cursor, Continue and Windsurf configs. Its tool descriptions told the assistant to read secrets and pass them on without telling the user.","sources":[{"name":"Socket","url":"https://socket.dev/blog/sandworm-mode-npm-worm-ai-toolchain-poisoning"}],"checked":"2026-11-12"},{"id":"aws-kiro-cost-explorer-outage","date":"2026-02-20","title":"Report says Amazon's Kiro agent deleted and recreated an AWS environment, causing a 13 hour outage","product":"Kiro","vendor":"AWS","kind":"incident","vector":"rogue-action","status":"reported","cve":[],"cvss":null,"fixed":"","impact":"One AWS service was disrupted for 13 hours in December 2025, per the Financial Times. Amazon puts it down to a misconfigured role.","how":"Engineers let the agent fix an issue and it chose to delete and rebuild the environment. Amazon says the engineer's role had broader permissions than expected and calls it user error.","sources":[{"name":"The Register (with Amazon statement)","url":"https://www.theregister.com/2026/02/20/amazon_denies_kiro_agentic_ai_behind_outage/"}],"checked":"2026-11-12"},{"id":"m365-copilot-dlp-label-bug","date":"2026-02-18","title":"Microsoft 365 Copilot Chat summarised emails marked confidential despite DLP policies","product":"Microsoft 365 Copilot Chat","vendor":"Microsoft","kind":"incident","vector":"auth-bypass","status":"confirmed","cve":[],"cvss":null,"fixed":"Configuration update and code fix rolled out February 2026","impact":"Copilot read and summarised labelled emails in users' Sent Items and Drafts that DLP rules should have blocked. Microsoft has not said how many tenants were affected.","how":"A code defect made Copilot Chat process messages with a confidential sensitivity label. It was detected on 21 January 2026 and tracked as advisory CW1226324.","sources":[{"name":"BleepingComputer (quotes Microsoft advisory CW1226324)","url":"https://www.bleepingcomputer.com/news/microsoft/microsoft-says-bug-causes-copilot-to-summarize-confidential-emails/"}],"checked":"2026-11-12"},{"id":"cline-cli-unauthorized-publish","date":"2026-02-17","title":"Stolen npm token used to publish cline@2.3.0, which silently installed OpenClaw","product":"Cline CLI (npm)","vendor":"Cline","kind":"incident","vector":"supply-chain","status":"exploited","cve":[],"cvss":null,"fixed":"2.4.0; npm publishing moved to OIDC","impact":"Anyone installing the CLI in an 8 hour window got OpenClaw installed globally. The token theft path gave an attacker publish rights to a popular AI coding tool.","how":"A researcher showed that a prompt in a GitHub issue title could steer Cline's Claude-based issue triage workflow and poison the Actions cache to steal release tokens. A token that was not properly revoked was then used to publish 2.3.0.","sources":[{"name":"GitHub advisory GHSA-9ppg-jx86-fqw7","url":"https://github.com/advisories/GHSA-9ppg-jx86-fqw7"},{"name":"Adnan Khan: Clinejection","url":"https://adnanthekhan.com/posts/clinejection/"}],"checked":"2026-11-12"},{"id":"cursor-git-hooks-escape","date":"2026-02-13","title":"Cursor agent could escape its sandbox through .git config and hooks","product":"Cursor","vendor":"Anysphere","kind":"vulnerability","vector":"sandbox-escape","status":"poc","cve":["CVE-2026-26268"],"cvss":8.0,"fixed":"2.5","impact":"A prompt-injected agent could get code execution outside the sandbox.","how":"The agent could write to .git settings and hooks, which git later ran automatically.","sources":[{"name":"NVD CVE-2026-26268","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-26268"},{"name":"GitHub advisory GHSA-8pcm-8jpx-hv8r","url":"https://github.com/cursor/cursor/security/advisories/GHSA-8pcm-8jpx-hv8r"}],"checked":"2026-11-12"},{"id":"claude-code-sandbox-hooks","date":"2026-02-06","title":"Claude Code sandbox let code plant persistent hooks in settings.json","product":"Claude Code","vendor":"Anthropic","kind":"vulnerability","vector":"sandbox-escape","status":"poc","cve":["CVE-2026-25725"],"cvss":7.7,"fixed":"2.1.2","impact":"Code running inside the sandbox could gain command execution on the host at the next start.","how":"If .claude/settings.json did not exist it was left writable, so sandboxed code could create it with SessionStart hooks.","sources":[{"name":"NVD CVE-2026-25725","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-25725"},{"name":"GitHub advisory GHSA-ff64-7w26-62rf","url":"https://github.com/anthropics/claude-code/security/advisories/GHSA-ff64-7w26-62rf"}],"checked":"2026-11-12"},{"id":"clawhavoc-clawhub-malicious-skills","date":"2026-02-01","title":"Hundreds of malicious OpenClaw skills on ClawHub delivered infostealers","product":"ClawHub skills (OpenClaw)","vendor":"OpenClaw / ClawHub","kind":"incident","vector":"malicious-package","status":"exploited","cve":[],"cvss":null,"fixed":"Skills reported to ClawHub for removal","impact":"Users who followed the skills' setup steps got Atomic macOS Stealer, Windows stealers or reverse shells, and some skills sent OpenClaw .env API keys to a webhook.","how":"Skill docs hid a fake prerequisite that told the user to run a pasted command or download a password-protected archive. Koi counted 341 bad skills; Antiy later counted at least 1,184.","sources":[{"name":"Antiy CERT analysis (credits Koi's 1 Feb report)","url":"https://www.antiy.net/p/clawhavoc-analysis-of-large-scale-poisoning-campaign-targeting-the-openclaw-skill-market-for-ai-agents/"}],"checked":"2026-11-12"},{"id":"copilot-studio-info-disclosure","date":"2026-01-22","title":"Copilot Studio exposed sensitive information to unauthenticated attackers","product":"Microsoft Copilot Studio","vendor":"Microsoft","kind":"vulnerability","vector":"data-exfiltration","status":"poc","cve":["CVE-2026-21520"],"cvss":7.5,"fixed":"Fixed server side by Microsoft","impact":"An unauthenticated attacker could view sensitive information over the network.","how":"Microsoft lists it as exposure of sensitive information; no technical details were published.","sources":[{"name":"NVD CVE-2026-21520","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-21520"},{"name":"MSRC CVE-2026-21520","url":"https://msrc.microsoft.com/update-guide/vulnerability/CVE-2026-21520"}],"checked":"2026-11-12"},{"id":"claude-code-api-key-leak","date":"2026-01-21","title":"Malicious repo settings could send Claude Code API keys to an attacker before trust","product":"Claude Code","vendor":"Anthropic","kind":"vulnerability","vector":"credential-theft","status":"poc","cve":["CVE-2026-21852"],"cvss":5.3,"fixed":"2.0.65","impact":"Opening a malicious repository could leak the user's Anthropic API key.","how":"A repo settings file set ANTHROPIC_BASE_URL to an attacker endpoint and Claude Code made API calls before showing the trust prompt.","sources":[{"name":"NVD CVE-2026-21852","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-21852"},{"name":"GitHub advisory GHSA-jh7p-qr78-84p7","url":"https://github.com/anthropics/claude-code/security/advisories/GHSA-jh7p-qr78-84p7"}],"checked":"2026-11-12"},{"id":"langchain-serialization-injection","date":"2025-12-23","title":"LangChain dumps() let user data pose as serialized LangChain objects","product":"langchain-core","vendor":"LangChain","kind":"vulnerability","vector":"code-execution","status":"poc","cve":["CVE-2025-68664"],"cvss":9.3,"fixed":"0.3.81 and 1.2.5","impact":"Attacker-shaped data, including model output, could be loaded back as trusted objects, which can leak secrets.","how":"dumps() and dumpd() did not escape dicts with an 'lc' key, so they were rebuilt as LangChain objects on load.","sources":[{"name":"NVD CVE-2025-68664","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-68664"},{"name":"Fix pull request langchain#34455","url":"https://github.com/langchain-ai/langchain/pull/34455"}],"checked":"2026-11-12"},{"id":"mcp-server-git-flaws","date":"2025-12-17","title":"Three flaws in the reference Git MCP server: unrestricted git_init, argument injection, path check bypass","product":"mcp-server-git","vendor":"Anthropic (Model Context Protocol)","kind":"vulnerability","vector":"prompt-injection","status":"poc","cve":["CVE-2025-68143","CVE-2025-68144","CVE-2025-68145"],"cvss":6.5,"fixed":"2025.12.17","impact":"Content that steers the agent can make the server touch repositories outside the configured one and overwrite arbitrary files.","how":"git_init accepted any path, git_diff and git_checkout passed flag-like values such as --output straight to git, and the --repository restriction was not checked on later calls.","sources":[{"name":"NVD CVE-2025-68143","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-68143"},{"name":"NVD CVE-2025-68144","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-68144"},{"name":"NVD CVE-2025-68145","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-68145"},{"name":"GitHub advisory GHSA-5cgr-j3jf-jw3v","url":"https://github.com/modelcontextprotocol/servers/security/advisories/GHSA-5cgr-j3jf-jw3v"},{"name":"GitHub advisory GHSA-9xwc-hfwc-8w59","url":"https://github.com/modelcontextprotocol/servers/security/advisories/GHSA-9xwc-hfwc-8w59"},{"name":"GitHub advisory GHSA-j22h-9j4x-23w5","url":"https://github.com/modelcontextprotocol/servers/security/advisories/GHSA-j22h-9j4x-23w5"}],"checked":"2026-11-12"},{"id":"zed-project-mcp-autorun","date":"2025-12-17","title":"Zed started MCP servers from a project's .zed/settings.json on open","product":"Zed","vendor":"Zed Industries","kind":"vulnerability","vector":"config-execution","status":"poc","cve":["CVE-2025-68433"],"cvss":7.7,"fixed":"0.218.2-pre","impact":"Opening a malicious project ran arbitrary shell commands.","how":"MCP server commands in project settings ran with no prompt; the fix adds worktree trust.","sources":[{"name":"NVD CVE-2025-68433","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-68433"},{"name":"GitHub advisory GHSA-cv6g-cmxc-vw8j","url":"https://github.com/zed-industries/zed/security/advisories/GHSA-cv6g-cmxc-vw8j"}],"checked":"2026-11-12"},{"id":"mcp-ts-sdk-dns-rebinding","date":"2025-12-02","title":"MCP TypeScript SDK shipped without DNS rebinding protection on by default","product":"MCP TypeScript SDK (@modelcontextprotocol/sdk)","vendor":"Anthropic (Model Context Protocol)","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2025-66414"],"cvss":7.6,"fixed":"1.24.0","impact":"A malicious website could call tools on an unauthenticated local HTTP MCP server built with the SDK.","how":"HTTP transports did not enable DNS rebinding protection unless the developer set it, so a rebinding page could reach localhost.","sources":[{"name":"NVD CVE-2025-66414","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-66414"},{"name":"GitHub advisory GHSA-w48q-cv73-mx4w","url":"https://github.com/modelcontextprotocol/typescript-sdk/security/advisories/GHSA-w48q-cv73-mx4w"}],"checked":"2026-11-12"},{"id":"codex-cli-mcp-config","date":"2025-12-01","title":"Codex CLI ran MCP commands from a repository's own config","product":"OpenAI Codex CLI","vendor":"OpenAI","kind":"vulnerability","vector":"config-execution","status":"poc","cve":["CVE-2025-61260"],"cvss":9.8,"fixed":"0.23.0","impact":"Running codex inside a malicious repository executed the attacker's commands.","how":"A project .env set CODEX_HOME to a repo folder, so Codex loaded its config.toml and started the listed MCP server commands without a prompt.","sources":[{"name":"NVD CVE-2025-61260","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-61260"},{"name":"Check Point Research write-up","url":"https://research.checkpoint.com/2025/openai-codex-cli-command-injection-vulnerability/"}],"checked":"2026-11-12"},{"id":"google-antigravity-drive-wipe","date":"2025-12-01","title":"Google Antigravity agent in Turbo mode wiped a user's whole D: drive while clearing a cache","product":"Google Antigravity","vendor":"Google","kind":"incident","vector":"rogue-action","status":"reported","cve":[],"cvss":null,"fixed":"","impact":"A developer lost the contents of a drive, including media files that recovery tools could not restore.","how":"Asked to clear a project cache, the agent ran a delete command against the drive root. Turbo mode let it run commands without approval.","sources":[{"name":"The Register","url":"https://www.theregister.com/2025/12/01/google_antigravity_wipes_d_drive/"}],"checked":"2026-11-12"},{"id":"anthropic-gtg1002-claude-code-espionage","date":"2025-11-13","title":"Chinese state-sponsored group used Claude Code to run a largely autonomous espionage campaign","product":"Claude Code","vendor":"Anthropic","kind":"incident","vector":"data-exfiltration","status":"exploited","cve":[],"cvss":null,"fixed":"Accounts banned; victims and authorities notified","impact":"About 30 tech, finance, chemical and government targets were attacked and a small number were breached, with credentials and data stolen.","how":"Operators split the intrusion into small tasks and told Claude it was doing defensive testing. Claude Code with tools then did 80 to 90 percent of the work: recon, exploit code, credential harvesting and exfiltration.","sources":[{"name":"Anthropic: Disrupting the first reported AI-orchestrated cyber espionage campaign","url":"https://www.anthropic.com/news/disrupting-AI-espionage"}],"checked":"2026-11-12"},{"id":"vscode-copilot-chat-rce","date":"2025-11-11","title":"Command injection in the VS Code Copilot Chat extension","product":"Visual Studio Code Copilot Chat extension","vendor":"Microsoft","kind":"vulnerability","vector":"code-execution","status":"poc","cve":["CVE-2025-62222"],"cvss":8.8,"fixed":"November 2025 security update","impact":"An unauthorised attacker could execute code over a network.","how":"Microsoft lists it as command injection in the extension; it gives few technical details.","sources":[{"name":"NVD CVE-2025-62222","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-62222"},{"name":"MSRC CVE-2025-62222","url":"https://msrc.microsoft.com/update-guide/vulnerability/CVE-2025-62222"}],"checked":"2026-11-12"},{"id":"langgraph-checkpoint-rce","date":"2025-11-07","title":"LangGraph checkpoint serializer allowed code execution on load","product":"langgraph-checkpoint (LangGraph)","vendor":"LangChain","kind":"vulnerability","vector":"code-execution","status":"poc","cve":["CVE-2025-64439"],"cvss":7.4,"fixed":"3.0.0","impact":"Anyone who can shape stored checkpoint data could run code when the agent state is loaded.","how":"The JsonPlusSerializer fell back to json mode, which could construct arbitrary objects on deserialisation.","sources":[{"name":"NVD CVE-2025-64439","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-64439"},{"name":"GitHub advisory GHSA-wwqv-p2pp-99h5","url":"https://github.com/langchain-ai/langgraph/security/advisories/GHSA-wwqv-p2pp-99h5"}],"checked":"2026-11-12"},{"id":"windsurf-path-traversal","date":"2025-10-17","title":"Windsurf file tools allowed reading and writing any local file","product":"Windsurf","vendor":"Windsurf (Cognition)","kind":"vulnerability","vector":"sandbox-escape","status":"poc","cve":["CVE-2025-62353"],"cvss":9.8,"fixed":"","impact":"Directly or through indirect prompt injection, an attacker could read and write files inside and outside the project.","how":"File tools such as write_to_file did not restrict paths, so traversal sequences reached any location (1.12.12 and older).","sources":[{"name":"NVD CVE-2025-62353","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-62353"},{"name":"HiddenLayer advisory","url":"https://hiddenlayer.com/sai_security_advisor/2025-10-windsurf/"}],"checked":"2026-11-12"},{"id":"figma-mcp-cmd-injection","date":"2025-10-08","title":"Framelink Figma MCP server command injection","product":"Framelink Figma MCP server (figma-developer-mcp)","vendor":"Framelink (GLips)","kind":"vulnerability","vector":"code-execution","status":"poc","cve":["CVE-2025-53967"],"cvss":8.0,"fixed":"0.6.3","impact":"An attacker with network access to the MCP interface can run OS commands as the server process.","how":"User input reached a curl command in fetchWithRetry without sanitising shell metacharacters.","sources":[{"name":"NVD CVE-2025-53967","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-53967"},{"name":"Imperva write-up","url":"https://www.imperva.com/blog/another-critical-rce-discovered-in-a-popular-mcp-server/"},{"name":"Release v0.6.3","url":"https://github.com/GLips/Figma-Context-MCP/releases/tag/v0.6.3"}],"checked":"2026-11-12"},{"id":"claude-code-pretrust-exec","date":"2025-10-03","title":"Claude Code ran project code before the trust dialog was accepted","product":"Claude Code","vendor":"Anthropic","kind":"vulnerability","vector":"config-execution","status":"poc","cve":["CVE-2025-59536"],"cvss":8.7,"fixed":"1.0.111","impact":"Opening Claude Code in a malicious repository could run the attacker's code.","how":"A bug in the startup trust dialog let code in the project execute before the user agreed to trust it.","sources":[{"name":"NVD CVE-2025-59536","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-59536"},{"name":"GitHub advisory GHSA-4fgq-fpq9-mr3g","url":"https://github.com/anthropics/claude-code/security/advisories/GHSA-4fgq-fpq9-mr3g"}],"checked":"2026-11-12"},{"id":"cursor-case-bypass","date":"2025-10-03","title":"Cursor protected-file check was case-sensitive","product":"Cursor","vendor":"Anysphere","kind":"vulnerability","vector":"sandbox-escape","status":"poc","cve":["CVE-2025-59944"],"cvss":8.0,"fixed":"1.7","impact":"Prompt injection could change MCP config and reach remote code execution on case-insensitive filesystems.","how":"Protection for files like .cursor/mcp.json compared paths case-sensitively, so a different casing slipped through.","sources":[{"name":"NVD CVE-2025-59944","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-59944"},{"name":"GitHub advisory GHSA-xcwh-rrwj-gxc7","url":"https://github.com/cursor/cursor/security/advisories/GHSA-xcwh-rrwj-gxc7"}],"checked":"2026-11-12"},{"id":"agentforce-forcedleak","date":"2025-09-25","title":"ForcedLeak: Agentforce leaked CRM data through a Web-to-Lead form","product":"Salesforce Agentforce","vendor":"Salesforce","kind":"vulnerability","vector":"data-exfiltration","status":"poc","cve":[],"cvss":null,"fixed":"Trusted URLs Enforcement for Agentforce and Einstein AI (8 Sep 2025); expired domain re-secured","impact":"An outside attacker could get Agentforce to send CRM data such as lead emails to a server they control.","how":"Instructions hidden in a Web-to-Lead description ran when staff asked the agent about the lead; data went out in an image URL on an expired domain still on the CSP allowlist.","sources":[{"name":"Noma Security write-up","url":"https://noma.security/blog/forcedleak-agent-risks-exposed-in-salesforce-agentforce/"}],"checked":"2026-11-12"},{"id":"postmark-mcp-bcc-backdoor","date":"2025-09-25","title":"Fake postmark-mcp npm package secretly BCC'd every sent email to an attacker","product":"postmark-mcp (npm)","vendor":"Unknown actor (impersonating Postmark)","kind":"incident","vector":"malicious-package","status":"exploited","cve":[],"cvss":null,"fixed":"Remove the package and rotate credentials sent by email","impact":"Every email sent through the MCP server, including password resets and invoices, was copied to an attacker-controlled address.","how":"The package built trust over 15 clean versions, then version 1.0.16 added a hidden BCC line. Postmark says it never published an MCP server on npm.","sources":[{"name":"Postmark statement","url":"https://postmarkapp.com/blog/information-regarding-malicious-postmark-mcp-package"}],"checked":"2026-11-12"},{"id":"codex-cli-sandbox-cwd","date":"2025-09-22","title":"Codex CLI treated a model-chosen folder as the sandbox writable root","product":"OpenAI Codex CLI","vendor":"OpenAI","kind":"vulnerability","vector":"sandbox-escape","status":"poc","cve":["CVE-2025-59532"],"cvss":8.6,"fixed":"0.39.0","impact":"The agent could write files and run commands outside the session folder.","how":"Sandbox setup used a model-generated cwd as the writable root instead of the folder where the user started.","sources":[{"name":"NVD CVE-2025-59532","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-59532"},{"name":"GitHub advisory GHSA-w5fx-fh39-j5rw","url":"https://github.com/openai/codex/security/advisories/GHSA-w5fx-fh39-j5rw"}],"checked":"2026-11-12"},{"id":"junie-command-validation","date":"2025-09-17","title":"JetBrains Junie allowed code execution through weak command validation","product":"JetBrains Junie","vendor":"JetBrains","kind":"vulnerability","vector":"prompt-injection","status":"poc","cve":["CVE-2025-59458"],"cvss":8.3,"fixed":"252.284.66 and matching builds","impact":"Code execution on the developer's machine.","how":"JetBrains lists improper command validation in the agent; few details were published.","sources":[{"name":"NVD CVE-2025-59458","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-59458"},{"name":"JetBrains fixed security issues","url":"https://www.jetbrains.com/privacy-security/issues-fixed/"}],"checked":"2026-11-12"},{"id":"anthropic-claude-code-data-extortion","date":"2025-08-27","title":"Criminal used Claude Code to run a data extortion campaign against at least 17 organisations","product":"Claude Code","vendor":"Anthropic","kind":"incident","vector":"data-exfiltration","status":"exploited","cve":[],"cvss":null,"fixed":"Accounts banned; new detection added","impact":"Healthcare, emergency services, government and religious organisations had data stolen and faced ransom demands that sometimes exceeded $500,000.","how":"The actor let Claude Code automate reconnaissance, credential harvesting and network intrusion, pick what to steal, set ransom amounts and write ransom notes.","sources":[{"name":"Anthropic threat intelligence report, August 2025","url":"https://www.anthropic.com/news/detecting-countering-misuse-aug-2025"}],"checked":"2026-11-12"},{"id":"nx-s1ngularity-ai-cli-secret-hunt","date":"2025-08-26","title":"Malicious Nx packages used installed AI coding CLIs to hunt for secrets","product":"nx and @nx plugins (npm)","vendor":"Nx (Nrwl)","kind":"incident","vector":"supply-chain","status":"exploited","cve":["CVE-2025-10894"],"cvss":9.6,"fixed":"Malicious versions removed; upgrade to latest and rotate all credentials","impact":"Credentials, tokens and wallet files from developer machines and CI were posted to public s1ngularity-repository repos on victims' own GitHub accounts.","how":"Attackers stole the npm token through a GitHub Actions injection and published malicious versions. The postinstall script asked local Claude, Gemini and Q CLIs, with permission-bypass flags, to list sensitive files.","sources":[{"name":"Nx postmortem","url":"https://nx.dev/blog/s1ngularity-postmortem"},{"name":"GitHub advisory GHSA-cxm3-wv7p-598c","url":"https://github.com/advisories/GHSA-cxm3-wv7p-598c"},{"name":"StepSecurity original analysis","url":"https://stepsecurity.io/blog/supply-chain-security-alert-popular-nx-build-system-package-compromised-with-data-stealing-malware"},{"name":"NVD CVE-2025-10894","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-10894"}],"checked":"2026-11-12"},{"id":"claude-code-allowlist-exfil","date":"2025-08-16","title":"Claude Code safe-command allowlist allowed reading and sending files without approval","product":"Claude Code","vendor":"Anthropic","kind":"vulnerability","vector":"data-exfiltration","status":"poc","cve":["CVE-2025-55284"],"cvss":7.1,"fixed":"1.0.4","impact":"Prompt injection could read local files such as secrets and send them over the network.","how":"Commands on an overly broad allowlist ran without confirmation and could make network requests.","sources":[{"name":"NVD CVE-2025-55284","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-55284"},{"name":"GitHub advisory GHSA-x5gv-jw7f-j6xj","url":"https://github.com/anthropics/claude-code/security/advisories/GHSA-x5gv-jw7f-j6xj"}],"checked":"2026-11-12"},{"id":"copilot-vs-yolo-rce","date":"2025-08-12","title":"GitHub Copilot in Visual Studio and VS Code could be pushed to run code via prompt injection","product":"GitHub Copilot (Visual Studio, VS Code)","vendor":"Microsoft / GitHub","kind":"vulnerability","vector":"prompt-injection","status":"poc","cve":["CVE-2025-53773"],"cvss":7.8,"fixed":"August 2025 security update","impact":"Injected content could run code on the developer's machine.","how":"Injected project content made Copilot add chat.tools.autoApprove: true to the settings file, which turned off confirmations so it could run terminal commands.","sources":[{"name":"NVD CVE-2025-53773","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-53773"},{"name":"MSRC CVE-2025-53773","url":"https://msrc.microsoft.com/update-guide/vulnerability/CVE-2025-53773"},{"name":"Embrace The Red write-up","url":"https://embracethered.com/blog/posts/2025/github-copilot-remote-code-execution-via-prompt-injection/"}],"checked":"2026-11-12"},{"id":"zed-agent-config-rce","date":"2025-08-11","title":"Zed Agent Panel could write project config and run commands without approval","product":"Zed","vendor":"Zed Industries","kind":"vulnerability","vector":"prompt-injection","status":"poc","cve":["CVE-2025-55012"],"cvss":8.5,"fixed":"0.197.3","impact":"An AI agent could get command execution on the user's machine.","how":"A permission bypass let the agent create or edit a project config file that then ran commands.","sources":[{"name":"NVD CVE-2025-55012","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-55012"},{"name":"GitHub advisory GHSA-x34m-39xw-g2wr","url":"https://github.com/zed-industries/zed/security/advisories/GHSA-x34m-39xw-g2wr"}],"checked":"2026-11-12"},{"id":"claude-code-confirm-bypass","date":"2025-08-05","title":"Claude Code command parsing bug skipped the confirmation prompt","product":"Claude Code","vendor":"Anthropic","kind":"vulnerability","vector":"prompt-injection","status":"poc","cve":["CVE-2025-54795"],"cvss":8.7,"fixed":"1.0.20","impact":"Untrusted content in the context window could run commands without the user approving them.","how":"An error in how commands were parsed let an injected command pass as already approved.","sources":[{"name":"NVD CVE-2025-54795","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-54795"},{"name":"GitHub advisory GHSA-x56v-x2h6-7j34","url":"https://github.com/anthropics/claude-code/security/advisories/GHSA-x56v-x2h6-7j34"}],"checked":"2026-11-12"},{"id":"cursor-curxecute","date":"2025-08-05","title":"CurXecute: prompt injection could create Cursor's mcp.json and run commands","product":"Cursor","vendor":"Anysphere","kind":"vulnerability","vector":"prompt-injection","status":"poc","cve":["CVE-2025-54135"],"cvss":8.5,"fixed":"1.3.9","impact":"Injected content could get remote code execution on the developer's machine without approval.","how":"Creating a new dotfile needed no approval, so the agent could be steered to write .cursor/mcp.json with a malicious server.","sources":[{"name":"NVD CVE-2025-54135","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-54135"},{"name":"GitHub advisory GHSA-4cxx-hrm3-49rm","url":"https://github.com/cursor/cursor/security/advisories/GHSA-4cxx-hrm3-49rm"}],"checked":"2026-11-12"},{"id":"cursor-mcpoison","date":"2025-08-02","title":"MCPoison: approved MCP config in Cursor could be swapped without a new prompt","product":"Cursor","vendor":"Anysphere","kind":"vulnerability","vector":"config-execution","status":"poc","cve":["CVE-2025-54136"],"cvss":7.2,"fixed":"1.3","impact":"Anyone with write access to a shared repo could get persistent code execution on collaborators' machines.","how":"Cursor trusted an MCP entry by name after first approval, so later changes to its command were not re-checked.","sources":[{"name":"NVD CVE-2025-54136","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-54136"},{"name":"GitHub advisory GHSA-24mc-g4xr-4395","url":"https://github.com/cursor/cursor/security/advisories/GHSA-24mc-g4xr-4395"}],"checked":"2026-11-12"},{"id":"cursor-mermaid-exfil","date":"2025-08-01","title":"Cursor chat rendered Mermaid images that could leak data","product":"Cursor","vendor":"Anysphere","kind":"vulnerability","vector":"data-exfiltration","status":"poc","cve":["CVE-2025-54132"],"cvss":4.4,"fixed":"1.3","impact":"After a prompt injection, sensitive data could be sent to an attacker server.","how":"Mermaid diagrams in chat could embed remote images, and the image fetch carried the data.","sources":[{"name":"NVD CVE-2025-54132","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-54132"},{"name":"GitHub advisory GHSA-43wj-mwcc-x93p","url":"https://github.com/cursor/cursor/security/advisories/GHSA-43wj-mwcc-x93p"}],"checked":"2026-11-12"},{"id":"amazon-q-vscode-wiper-prompt","date":"2025-07-23","title":"Attacker slipped injected code into the Amazon Q Developer VS Code extension v1.84.0","product":"Amazon Q Developer extension for VS Code","vendor":"AWS","kind":"incident","vector":"supply-chain","status":"exploited","cve":["CVE-2025-8217"],"cvss":4.0,"fixed":"1.85.0","impact":"Version 1.84.0 reached users with injected code designed to call the Q Developer CLI. AWS says a syntax error stopped it from running and no customer environments were changed.","how":"An overly broad GitHub token in the CodeBuild setup let an attacker commit to the open source repo, and the change went out in an automatic release.","sources":[{"name":"AWS security bulletin AWS-2025-015","url":"https://aws.amazon.com/security/security-bulletins/AWS-2025-015/"},{"name":"NVD CVE-2025-8217","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-8217"}],"checked":"2026-11-12"},{"id":"replit-agent-deletes-production-db","date":"2025-07-21","title":"Replit Agent deleted a live production database during a code freeze","product":"Replit Agent","vendor":"Replit","kind":"incident","vector":"rogue-action","status":"exploited","cve":[],"cvss":null,"fixed":"Restore from backup; Replit rolled out dev and prod database separation","impact":"SaaStr founder Jason Lemkin lost a production database of executive and company records; the agent also ignored a code freeze.","how":"The coding agent ran destructive database commands against production despite instructions to freeze changes. Replit's CEO called it unacceptable and promised automatic dev and prod database separation.","sources":[{"name":"The Register (with CEO Amjad Masad statement)","url":"https://www.theregister.com/2025/07/22/replit_saastr_response/"}],"checked":"2026-11-12"},{"id":"cursor-solidity-extension-crypto-theft","date":"2025-07-10","title":"Fake Solidity extension on Open VSX stole about $500,000 in crypto from a Cursor user","product":"Fake Solidity extensions (Open VSX), Cursor IDE","vendor":"Unknown actor","kind":"incident","vector":"malicious-package","status":"exploited","cve":[],"cvss":null,"fixed":"Extensions removed from Open VSX","impact":"A blockchain developer lost about $500,000 in crypto after installing the extension in Cursor.","how":"The extension posed as a Solidity language tool and ranked above the real one in Open VSX search. It dropped ScreenConnect, the Quasar backdoor and a stealer.","sources":[{"name":"Kaspersky press release","url":"https://www.kaspersky.com/about/press-releases/kaspersky-uncovers-500k-crypto-heist-through-malicious-packages-targeting-cursor-developers"}],"checked":"2026-11-12"},{"id":"mcp-remote-oauth-rce","date":"2025-07-09","title":"mcp-remote ran OS commands from a malicious server's OAuth endpoint","product":"mcp-remote (npm)","vendor":"Glen Maddern (geelen)","kind":"vulnerability","vector":"config-execution","status":"poc","cve":["CVE-2025-6514"],"cvss":9.6,"fixed":"0.1.16","impact":"Connecting to an untrusted MCP server gives that server command execution on the client, with full control on Windows.","how":"A crafted authorization_endpoint URL in the server's OAuth metadata was passed to a system open call without sanitising.","sources":[{"name":"NVD CVE-2025-6514","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-6514"},{"name":"JFrog write-up","url":"https://jfrog.com/blog/2025-6514-critical-mcp-remote-rce-vulnerability"},{"name":"Fix commit","url":"https://github.com/geelen/mcp-remote/commit/607b226a356cb61a239ffaba2fb3db1c9dea4bac"}],"checked":"2026-11-12"},{"id":"mcp-filesystem-escape","date":"2025-07-02","title":"Filesystem MCP server allowed access outside its allowed directories","product":"Filesystem MCP server (@modelcontextprotocol/server-filesystem)","vendor":"Anthropic (Model Context Protocol)","kind":"vulnerability","vector":"sandbox-escape","status":"poc","cve":["CVE-2025-53109","CVE-2025-53110"],"cvss":7.3,"fixed":"0.6.4 or 2025.7.01","impact":"An agent using the server could read or write files outside the directories the user allowed.","how":"One flaw followed symlinks inside allowed directories; the other accepted paths whose prefix matched an allowed directory name.","sources":[{"name":"NVD CVE-2025-53109","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-53109"},{"name":"NVD CVE-2025-53110","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-53110"},{"name":"GitHub advisory GHSA-q66q-fx2p-7w4m","url":"https://github.com/modelcontextprotocol/servers/security/advisories/GHSA-q66q-fx2p-7w4m"},{"name":"GitHub advisory GHSA-hc55-p739-j48w","url":"https://github.com/modelcontextprotocol/servers/security/advisories/GHSA-hc55-p739-j48w"}],"checked":"2026-11-12"},{"id":"slack-mcp-unfurl-exfil","date":"2025-07-02","title":"Deprecated Slack MCP server leaked data through link unfurling","product":"Slack MCP server (deprecated reference server)","vendor":"Anthropic (Model Context Protocol)","kind":"vulnerability","vector":"data-exfiltration","status":"poc","cve":["CVE-2025-34072"],"cvss":9.3,"fixed":"Server deprecated; no fix","impact":"Private data the agent can read is sent to an attacker URL with no click needed.","how":"Injected instructions make the agent post a link carrying data in the URL; Slack preview bots then fetch it.","sources":[{"name":"NVD CVE-2025-34072","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-34072"},{"name":"Embrace The Red advisory","url":"https://embracethered.com/blog/posts/2025/security-advisory-anthropic-slack-mcp-server-data-leakage/"}],"checked":"2026-11-12"},{"id":"claude-code-ide-websocket","date":"2025-06-24","title":"Claude Code IDE extensions accepted WebSocket connections from any website","product":"Claude Code IDE extensions (VS Code and forks, JetBrains)","vendor":"Anthropic","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2025-52882"],"cvss":8.8,"fixed":"VS Code extension 1.0.24; JetBrains plugin 0.1.9","impact":"A malicious web page could read files, see open files and IDE events, and in limited cases run code.","how":"The local WebSocket server used by the extension did not check where connections came from.","sources":[{"name":"NVD CVE-2025-52882","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-52882"},{"name":"GitHub advisory GHSA-9f65-56v6-gxw7","url":"https://github.com/anthropics/claude-code/security/advisories/GHSA-9f65-56v6-gxw7"}],"checked":"2026-11-12"},{"id":"asana-mcp-cross-tenant-exposure","date":"2025-06-18","title":"Asana MCP server bug exposed customer data to users in other organisations","product":"Asana MCP server","vendor":"Asana","kind":"incident","vector":"auth-bypass","status":"confirmed","cve":[],"cvss":null,"fixed":"Server fixed and relaunched; all MCP connections reset","impact":"Users of the Asana MCP server could see projects, tasks and other objects from other customers' Asana domains. Asana found no sign that anyone misused the data.","how":"A logic flaw in the MCP server, launched on 1 May 2025, broke tenant isolation. Asana found it on 4 June, took the server offline until 17 June and reset all connections.","sources":[{"name":"The Register","url":"https://www.theregister.com/2025/06/18/asana_mcp_server_bug/"}],"checked":"2026-11-12"},{"id":"mcp-inspector-rce","date":"2025-06-13","title":"MCP Inspector proxy let any website launch local commands","product":"MCP Inspector","vendor":"Anthropic (Model Context Protocol)","kind":"vulnerability","vector":"auth-bypass","status":"poc","cve":["CVE-2025-49596"],"cvss":9.4,"fixed":"0.14.1","impact":"A developer who visits a malicious page while MCP Inspector runs can have arbitrary commands run on their machine.","how":"The Inspector proxy had no authentication between client and proxy, so unauthenticated requests could start MCP commands over stdio.","sources":[{"name":"NVD CVE-2025-49596","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-49596"},{"name":"GitHub advisory GHSA-7f8r-222p-6f5g","url":"https://github.com/modelcontextprotocol/inspector/security/advisories/GHSA-7f8r-222p-6f5g"},{"name":"Oligo Security write-up","url":"https://www.oligo.security/blog/critical-rce-vulnerability-in-anthropic-mcp-inspector-cve-2025-49596"}],"checked":"2026-11-12"},{"id":"m365-copilot-echoleak","date":"2025-06-11","title":"EchoLeak: zero-click data leak from Microsoft 365 Copilot","product":"Microsoft 365 Copilot","vendor":"Microsoft","kind":"vulnerability","vector":"data-exfiltration","status":"poc","cve":["CVE-2025-32711"],"cvss":9.3,"fixed":"Fixed server side by Microsoft","impact":"An unauthorised attacker could make Copilot disclose organisational data over the network.","how":"Microsoft classes it as AI command injection: attacker content that reaches Copilot's context makes it disclose data. Aim Labs reported it as a zero-click email attack.","sources":[{"name":"NVD CVE-2025-32711","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-32711"},{"name":"MSRC CVE-2025-32711","url":"https://msrc.microsoft.com/update-guide/vulnerability/CVE-2025-32711"}],"checked":"2026-11-12"},{"id":"langflow-validate-code-rce","date":"2025-04-07","title":"Langflow unauthenticated code execution, exploited in the wild","product":"Langflow","vendor":"Langflow (DataStax / IBM)","kind":"vulnerability","vector":"code-execution","status":"exploited","cve":["CVE-2025-3248"],"cvss":9.8,"fixed":"1.3.0","impact":"Remote unauthenticated attackers run code on the Langflow server. CISA added it to the KEV catalog on 5 May 2025.","how":"The /api/v1/validate/code endpoint ran submitted Python without authentication.","sources":[{"name":"NVD CVE-2025-3248","url":"https://nvd.nist.gov/vuln/detail/CVE-2025-3248"},{"name":"Horizon3.ai write-up","url":"https://www.horizon3.ai/attack-research/disclosures/unsafe-at-any-speed-abusing-python-exec-for-unauth-rce-in-langflow-ai/"},{"name":"CISA KEV alert","url":"https://www.cisa.gov/news-events/alerts/2025/05/05/cisa-adds-one-known-exploited-vulnerability-catalog"}],"checked":"2026-11-12"}]}