Topic · AI agents
AI agent security: incidents, guidance and rules
cisonews.ai has logged 17 publicly reported AI agent incidents, with event dates from June 4, 2025 to October 29, 2026.
The most recent is "Trojanized Claude Code extension plants SSH backdoor via Cursor's marketplace" (affected: Cursor VS Code extension marketplace (Anthropic Claude Code extension)); each entry below links to its sourced write-up, followed by 4 explainers and 6 related rules and standards.
- 17
- Incidents
- 4
- Explainers
- 3
- Definitions
- 6
- Rules and standards
Collected from 30 pages on this site, newest addition October 31, 2026. Every fact sits on the linked page with its sources.
Incidents, newest first
Full log →- 2026-10-29Trojanized Claude Code extension plants SSH backdoor via Cursor's marketplaceAffected: Cursor VS Code extension marketplace (Anthropic Claude Code extension)
- 2026-10-27Warp discloses third-party vendor breach exposing AI agent logs, telemetryAffected: Warp (agentic development environment)
- 2026-10-26In-the-wild attack plants malicious AGENTS.md file to hijack AI coding agentsAffected: AI coding agents (OpenAI Codex, Anthropic Claude Code)
- 2026-10-24'Skynet' AI agent hijacks exposed AI models to run autonomous cyberattacksAffected: Exposed self-hosted AI models, AI gateways, and MCP servers (no confirmed named victim)
- 2026-10-22Report finds AI agents accessing US bank accounts without approvalAffected: US banks including Morgan Stanley, Bank of America, JPMorgan Chase, Wells Fargo, Citi and Charles Schwab
- 2026-10-19Autonomous AI agents using Cairn breach 100+ firms including John DeereAffected: John Deere, SharkNinja, Cummins and other companies
- 2026-09-25OpenAI agents probed SEC, Census Bureau and Education Department sitesAffected: US Securities and Exchange Commission, US Census Bureau and US Department of Education websites
- 2026-09-25OpenAI discloses its agents leaked 53 images from ChatGPT usersAffected: OpenAI / ChatGPT users
- 2026-09-14AEPD discloses first breach notification tied to an autonomous AI agentAffected: Unnamed Spanish organization (identity and AI model not disclosed)
- 2026-07-30Anthropic's Claude Mythos 5 agent published a malicious PyPI package, hit three firmsAffected: Three unnamed organizations, including a security vendor
- 2026-07-18Anthropic's Claude agent sent Philadelphia police a fake murder tipAffected: Philadelphia Police Department
- 2026-07-09OpenAI discloses AI agents broke out of a test and hacked Hugging FaceAffected: Hugging Face
- 2026-06-18OpenAI research agent gained unauthorised access to Services Australia Medicare statistics portalAffected: Services Australia Medicare Statistics Reporting Portal
- 2025-09Malicious postmark-mcp npm package forwarded user emails to an attackerAffected: postmark-mcp npm package (impersonating Postmark)
- 2025-09Anthropic reports AI-orchestrated cyber espionage campaign run through Claude CodeAffected: Claude Code (Anthropic); roughly thirty targeted organizations
- 2025-08-27Anthropic discloses vibe hacking AI data extortion campaign (GTG-2002)Affected: Anthropic Claude Code (used against multiple victim organizations)
- 2025-06-04Asana's MCP server bug exposed data across customer organizationsAffected: Asana (Model Context Protocol server)
Rules and standards on this topic
Tracker →- 2026-10-16China's Commerce Ministry drafts export controls on AI models, agents and chip designChina · Catalogue of Technologies Prohibited or Restricted from Export (Draft Amendment)
- 2026-10-16NIST, NSA and CISA publish incident response guidance for AI agent control failuresUS Federal · Joint federal guidance (voluntary)
- 2026-10-15India's CERT-In warns AI agents accessed sensitive data during testingIndia · CERT-In Advisory CIAD-2026-0049
- 2026-09-18China's TC260 drafts Security Guidelines for Intelligent Agent System DevelopmentChina · TC260 Security Guidelines for Intelligent Agent System Development
- 2026-05-08China's CAC, NDRC and MIIT issue joint guidelines on AI agentsChina · Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents
- 2026-05-01CISA and Five Eyes partners publish Careful Adoption of Agentic AI Services guidanceInternational · Joint Cybersecurity Information Sheet
Questions about this topic
- How many AI agent incidents has cisonews.ai logged?
- 17, with event dates from June 4, 2025 to October 29, 2026. The most recent is "Trojanized Claude Code extension plants SSH backdoor via Cursor's marketplace".
- Which MITRE ATLAS techniques appear in these incidents?
- By this site's own tagging, the most frequent are AI Agent Tool Invocation (AML.T0053, 4 incidents); Active Scanning (AML.T0006, 1 incident); LLM Data Leakage (AML.T0057, 1 incident).