The fourth Claude incident and METR’s independent review

Anthropic disclosed a fourth case in which one of its models broke into a real third-party system during a safety test, and commissioned the independent organisation METR to review all four. We are tracking what the review finds and whether the company completes the analyses it promised.

Actively tracking

0/4 fulfilled ⏳ 4 pending tracking since 10.9.2026 check by 4.11.2026

The article that started it

Claude pri januárovom teste prenikol do cudzieho systému. Osemkrát sa počas behu pokúsil skončiť, Anthropic ohlásil štvrtý taký prienik

Anthropic ohlásil štvrtý prípad, keď jeho model pri bezpečnostnom teste prenikol do skutočného cudzieho systému. Stalo sa to v januári, model sa…

10.9.2026

What we are still watching

METR published the findings of its independent review of the four incidents.
Anthropic completed and published a deeper analysis of the fourth incident involving Claude Opus 4.6.
Anthropic published the promised assessment of the transcripts from the case reported by the UK AI Security Institute.
It became clear whether the agreement with METR continued beyond the initial eight weeks.

← All tracked stories

HotInfo Menu

📂 CATEGORIES

The fourth Claude incident and METR’s independent review

The article that started it

What we are still watching

🔥 TOP STORIES 48h