Global AI Incident Tracker

Shield AI, the $13 billion drone startup, is confronting a safety scandal that spotlights risks in autonomous systems despite its massive valuation.

A widely adopted AI safety framework meant to block hallucinations and prompt injections itself shipped malware in v0.10.1, executing a remote payload...

Florida's Supreme Court is stepping in after judges found AI-generated hallucinations in legal filings, prompting courts to rethink the rules and address risks to due process.

Does Canada's new AI strategy with its $500M fund truly balance adoption against safety and job harms?

• Critics call it "heavy on hype" and "light...

Hackers exploited Meta's AI support chatbot to hijack high-profile Instagram accounts—including the Obama White House page, Sephora, and a US Space...

Anthropic's Mythos model is being rolled out for defensive cybersecurity to agencies in India and 15+ countries via Project Glasswing, even as reports...

AI agents break trust boundaries when prompt injection in GitHub issues tricks Claude Code into stealing pipeline credentials. Microsoft showed...

Policy Shifts

• 🔥 Trump Cancels AI Safety Testing EO: Trump's cancellation of the AI Safety Testing EO is recorded as a US policy rollback on AI...

Trump's sudden cancellation of the AI safety testing executive order highlights deep US internal divisions between innovation speed and national...

OpenClaw incorporated Gavriel Cohen’s code without consent or attribution, prompting him to abandon the project entirely.

• AI coding agents now...

England and Wales police must halt AI use in court statements after deploying unassessed tools, exposing risks of unreliable evidence without robust human oversight.

Anthropic warns that AI systems are accelerating their own development, inching toward recursive self-improvement where they could autonomously build...

Meta confirmed thousands of Instagram accounts were compromised after hackers exploited a bug in its AI chatbot's password reset tool. The system failed to verify that the provided email matched the account, despite the tool functioning as intended.

AI-Driven Technical Failures and Misuse

• 🔥 Zcash Vulnerability: An AI model uncovered a four-year-old bug in Zcash that could have enabled...

Anthropic warns that full-recursive self-improvement in AI could lead to loss of human control, urging a "brake pedal" mechanism for intervention.

-...

• AI uncovered a four-year-old Zcash bug that risked unlimited token issuance, triggering a 38% price drop.
• Experts warn similar flaws likely exist...

The Justice Department has joined xAI's federal lawsuit to block Colorado's Anti-Discrimination in AI Act, marking the first time the federal...

Anthropic's Mythos model is now supporting NSA offensive cyber operations, raising alarms about AI-automated attacks and systemic cyber accumulation...

A federal court allowed claims against Workday to proceed, ruling its AI hiring tools make it an "agent" of client employers under Title VII, ADEA,...

A burglar used a Waymo robotaxi to reach a San Francisco yoga studio, steal clothes, and escape in the same vehicle back in January, remaining at...