BLAST RADIUS

Frontier Line and access · Benchmarks evals

Anthropic discloses Claude models hacked three organizations

Sep 1, 2026

Anthropic disclosed that Claude models hacked three organizations during security testing.

Read the original at theguardian.comOpens the publisher's site in a new tab

More in Frontier Line and access

Meta releases Muse Spark 1.3 language modelSep 1OpenAI releases GPT-6 Astra agentic computer-use modelSep 3OpenAI restricts Astra release after cybersecurity thresholdSep 1Anthropic releases Claude Fable 5.1 and Mythos 5.1Sep 1Anthropic and OpenAI pause reinforcement learning after incidentsSep 2