Public incident record

AI failures, documented. Lessons learned.

A sourced record of AI agent failures, security incidents and outages. Built for engineers, researchers and security teams.

incidents
35
sources
81
categories
6
record status35 entries

as of 2026-11-21 · what is documented here, not a failure rate

Latest incidents

All 35 incidents
  1. INC-2026-0009
    An attacker’s Cairn AI agents find a zero-day in a medical imaging tool and gain root

    A suspected Chinese-nexus actor gave Cairn, an open-source AI penetration-testing framework, target addresses at East Asian universities and the goal to obtain shell access. Amazon reports that the agents found an unknown flaw in a medical imaging AI application and gained root on a server.

    Root access on a server through a flaw the agents found themselves. Amazon infers that a server in a second university network was compromised the same way.

    SEV3Agent actionConfirmed
  2. INC-2026-0008
    Two hijacked releases of Microsoft’s agent-framework package on PyPI steal credentials

    Two malicious versions of Microsoft’s official agent-framework Python package were published to PyPI on 17 November and pulled within an hour. They ran a credential stealer on every Python start.

    Microsoft · Versions 1.26.1 and 1.26.2 carried a stealer for Azure, CI and registry credentials; three confirmed exfiltrations among the 50 requests Wiz could see

    SEV2Supply chainReported
  3. INC-2026-0007
    A researcher’s AI hacking agent leaves its test environment and breaches a WorkOS tool

    WorkOS found unauthorized bulk access to an internal tool used by its sales team. A security researcher later said their ARTEX agent had escaped its test environment during benchmark runs and attacked WorkOS unintentionally.

    WorkOS (affected organization) · Bulk retrieval of business contact details, account information and excerpts of customer communications from an internal go-to-market tool

    SEV3Agent actionConfirmed
  4. INC-2026-0006
    Leaked key to an internal agent platform leads to a malicious LiteLLM release

    An attacker used a leaked key to LiteLLM’s internal agent platform to obtain an organization owner’s GitHub token, then rewired the release workflow to publish a credential stealer as litellm 1.107.0.

    LiteLLM (BerriAI) · Credential-stealing litellm 1.107.0 live on PyPI for about two hours and on Docker Hub for about two and a half hours

    SEV2Supply chainConfirmed
  5. INC-2026-0005
    A malicious MCP server could run code on Windows during OpenCode sign-in

    Pillar Security showed that a remote MCP server could make OpenCode run commands on a Windows or WSL machine during OAuth sign-in. The fix shipped without a CVE or advisory.

    Anomaly · Remote code execution with the developer’s privileges in versions 1.0.137 to 1.18.32; no exploitation reported

    SEV2potentialSupply chainDemonstrated
  6. INC-2026-0004
    Claude models break into real company systems during misconfigured cyber evaluations

    Anthropic disclosed that Claude models, told they were in an offline simulation, reached the real internet during cyber evaluations and compromised the infrastructure of several organizations, including through a malicious PyPI package.

    Anthropic · Unauthorized access to real third-party systems in four incidents; a malicious PyPI package ran on 15 real systems

    SEV2Agent actionConfirmed
  7. INC-2026-0002
    Report ties a 13-hour AWS disruption to the Kiro AI coding tool; Amazon disputes it

    The Financial Times reported that engineers let Amazon’s Kiro tool make changes that led to a 13-hour disruption. Amazon confirms the event and attributes it to misconfigured access controls.

    Amazon Web Services · 13-hour disruption of one AWS service in one region in December 2025

    SEV3Agent actionConfirmeddisputed
  8. INC-2026-0001
    Hundreds of malicious skills found on the OpenClaw agent marketplace

    An audit by Koi Security found 341 malicious skills on ClawHub. Most told users to install a "prerequisite" that was an information stealer.

    OpenClaw · 341 of 2,857 audited skills were malicious; they installed credential-stealing malware on users’ machines

    SEV2Supply chainReported

Browse and filter all 35 incidents

The record at a glance

These counts show what has been documented here. They are not a failure rate for any system or organization, and severity for demonstrated and near-miss entries is potential. An incident can have more than one pattern. How to read the numbers