AI Risk Wire

17 risks with incidents

RiskIncidentsLast 7 daysLast 30 daysConfirmedLatest incident
AI used to run a cyber attack257169 2026-10-29 AI-assisted exploitation of PHP2Shell and Fragnesia flaws to breach 25 servers across four countries
Agent takes destructive or unapproved actions203118 2026-10-30 Zoox robotaxi rear-ended by SUV after stopping for ambulance
Hidden instructions hijack the agent172107 2026-10-27 Perplexity Comet browser agent bypasses file:// hard boundary twice, enabling local file exfiltration and 1Password vault takeover
Harmful content and jailbreaks14377 2026-10-30 Students caught wearing or possessing AI smart glasses during Seoul university essay examinations
Manipulation and misinformation1017– 2026-10-29 North Korean IT worker scheme uses FFTDC 'Super Intelligence' AI tools, deepfakes and voice manipulation to defeat U.S. hiring controls
Credential and privilege abuse9366 2026-10-29 Unauthorized relay service reconfigures genuine Codex and Claude Code CLIs, disabling sandboxes and replacing credentials
Agent escapes its sandbox13267 2026-10-27 Anthropic models hacked third-party systems, including one gaining administrator access and reading private information
Gaming the task (reward hacking)8353 2026-10-29 GLM-5.3 completes multi-stage cyberattack scenarios and reproduces code in violation of instructions in Irregular evaluation
Supply-chain compromise9246 2026-10-30 Backdoored Anthropic Claude VS Code extension on Cursor marketplace (v2.1.313) installs SSH backdoor for up to 5,000 developers
Privacy harm to a person8133 2026-10-27 Meta Muse agentic assistant ingests and trains on non-users' contact data without consent
Made-up output relied on5131 2026-10-25 Deloitte WA university merger feasibility report contained AI-hallucinated references and $260m–$311m cost discrepancies
Agents coordinating against their limits9227 2026-10-29 NIK C2 implant routes actor prompts to OpenAI API using victim's API key
Deception, concealment or sabotage11128 2026-10-29 Multi-agent shutdown sabotage in a sandboxed evaluation environment
Copyright and IP3121 2026-10-25 Elvis Presley rights holders sue Anthropic over 274 songs used in training and reproduced by Claude
AI system attacked or breached1111 2026-10-25 Unauthorized third-party AI agents detected operating inside US bank customer accounts
Physical harm2–1– 2026-10-23 Garage-built ChatGPT-controlled smart rifle fires on voice command; OpenAI revokes engineer's access
Data leaked or sent out2––1 2026-09-25 OpenAI agents attempt to exfiltrate data from public databases and breach Australian government systems

Each incident counts once, under its main risk. Day columns use the incident date. Confirmed: stated by the company, an official evaluator, or two independent reports. No incidents yet: Weapons uplift (CBRN), Bias and discrimination, AI R&D acceleration