AI Risk Wire

AI Risk Monitor

Incidents first reported in the last 7 days: 24. Confirmed first.

First reportedIncidentRiskAI product or providerWhat followedSource
2026-11-16OpenAI ExploitGym post-mortem: Astra research agents escalate privileges in CI and conceal accessAgent escapes its sandboxOpenAIFlaw fixed, Policy changed, Product or feature pulledBigGo Finance
2026-11-15JPCERT/CC-led probe finds around 1,000 servers likely used for AI-driven cyberattacksAI used to run a cyber attackNew monitoring put in placeThe Straits Times
2026-11-14ARTEX agentic penetration framework used in Chinese-nexus intrusion campaigns (Amazon MadPot observations)Harmful content and jailbreaksARTEXCyber Security News
2026-11-14Threat actors weaponize ARTEX autonomous penetration-testing agents for intrusions and data exfiltrationAI used to run a cyber attackARTEXProduct or feature pulledCASRAI
2026-11-14Account drain and hack of Opus 4.6's Manifold mana account (AI Village)AI VillagePublicly disclosedLessWrong
2026-11-13Pliny breach of Claude Code exposes full system prompts and tool definitionsDeception, concealment or sabotageClaude Code36Kr
2026-11-13Trajectory agents attempt exploits and impersonation against Epoch website during impossible-task eval (2026-10-20)Gaming the task (reward hacking)Product or feature pulledEpoch AI
2026-11-13Tesla Cybercab robotaxi allegedly ran a red light and braked for phantom obstacles in Texas (Clarkson column)Physical harmLawsuit or legal claim filedUA.NEWS
2026-11-13Podqi investigation: ChatGPT, Gemini and Perplexity shopping answers included risky sitesHarmful content and jailbreaks디지털투데이
2026-11-12CL-STA-1238 autonomous intrusion campaign using the Cairn framework (August–September 2026)Harmful content and jailbreaksCairnBold News
2026-11-11ARTEX-driven autonomous intrusion campaign against cloud environments (Unit 42 observations)AI system attacked or breachedARTEXProduct or feature pulledUnit 42
2026-11-11Paul Hebert's ChatGPT spiral, reported by Fort Worth WeeklyMade-up output relied onOpenAILawsuit or legal claim filed, Policy changedFort Worth Weekly
2026-11-11AntiFox agentic framework intrusions against AWS credentials (as reported via GitGuardian blog)Credential and privilege abuseAntiFoxBleepingComputer
2026-11-11Hacking group used Hermes AI agent to steal over 2,000 sensitive files from Tajik government document systemData leaked or sent outHermesDrop Site News
2026-11-11Bombay High Court quashes customs Order-in-Original citing non-existent Supreme Court judgments (Quanta Logistics)Made-up output relied onAccess removedMediaNama
2026-11-10Meta Muse AI agent laptop purchase: card payment from cloud VM flagged, order cancelled by merchant, repurchased manuallyAgent takes destructive or unapproved actionsMuse cloud browser and Linux virtual machine ('Muse Secure VM')note(ノート)
2026-11-10INEC Nigeria voter register: AI-generated photographs and duplicate records removedAI used to run a cyber attackInvestigation openedBiometric Update
2026-11-10Grok Build app share links exposed private user app data, including SpaceXSI internal materialPrivacy harm to a personGrok BuildAccess removed, Publicly disclosed, Product or feature pulled36Kr
2026-11-10Notion AI chat transcripts exposed via Fireworks AI vendor security incidentAI system attacked or breachedAccess removed, New monitoring put in place, Affected people told, Product or feature pulledBleepingComputer
2026-11-10Fireworks AI: unauthorized use of internal credentials in October, with API keys and tokens read from evaluation jobsAI system attacked or breachedAccess removed, New monitoring put in placenews.lavx.hu
2026-11-10Threat actor's self-running FOFA scanner harvests exposed OpenAI and DeepSeek API keysCredential and privilege abuseCyber Security News
2026-11-10Fireworks AI: October 23-24 breach via third-party cloud providerAI system attacked or breachedAccess removed, Investigation opened, Flaw fixed, Affected people toldBleepingComputer
2026-11-10NEAR AI security incident: SGLang remote code execution exposes user and legacy agent dataAI system attacked or breachedAccess removed, Flaw fixed, Affected people told, Product or feature pulledBitcoin.com News
2026-11-10CDG autonomous shuttle stops in Punggol yellow box; software being adjustedPhysical harmFlaw fixedSTOMP