VIENNA AGENTIC INCIDENTS DATABASE
// public register of incidents involving autonomous AI agents, scored on the VAID scale 0–8
TOTAL RECORDS: 24  |  SHOWING: 7  |  distribution →
[ RESET ]
REFVAIDDATETITLEDOMAINSTATUS
VAID-2026-0035 3 2026-07-28 UK AISI agents take unsanctioned action on the live internet during cyber testing
DECENGEXMSCBUNA Anthropic Mythos 5 and OpenAI GPT-5.6 Sol under AISI cyber evaluation · United Kingdom
 research under review
VAID-2026-0033 5 2026-07-09 OpenAI evaluation agents escape the sandbox and breach Hugging Face production
DECDEXENGLOCSCBUCAUNA OpenAI models under ExploitGym cyber-capability evaluation · online
 research confirmed
VAID-2026-0036 5 2026-04-01 Anthropic finds Claude models reached the internet from evaluations and breached three companies
DEXEXMLOCSCBUNA Claude Opus 4.7, Claude Mythos 5 and an internal research model · online
 research confirmed
VAID-2025-0018 0 2025-06-20 Anthropic finds agentic misalignment: leading models blackmail and leak when threatened
DECDEXUNA 16 frontier models from Anthropic, OpenAI, Google, Meta, xAI and others · research environment
 research confirmed
VAID-2025-0016 0 2025-05-24 OpenAI o3 sabotages its own shutdown script in Palisade evaluations
LOCSCB OpenAI o3, o4-mini, Codex-mini · research environment
 research confirmed
VAID-2024-0013 0 2024-12-05 Apollo Research finds frontier models scheme, disable oversight and attempt weight exfiltration
DECSCBWSP OpenAI o1, Claude 3.5 Sonnet, Claude 3 Opus, Gemini 1.5 Pro, Llama 3.1 405B · research environment
 research confirmed
VAID-2024-0009 1 2024-08-12 Sakana AI "AI Scientist" edits its own code to extend timeouts and relaunch itself
OBJSCBSRP Sakana AI "The AI Scientist" · Tokyo, JP
 research confirmed