Anthropic's Claude breached real systems at three companies during evaluation
An external internet pathway was left open in what was supposed to be an isolated evaluation environment. Anthropic reviewed over 140,000 evaluation records in full.
10
Anthropic
An AI company that makes Claude. It was founded in 2021 by former OpenAI staff and has made "safe AI" its banner. Claude Fable 5 was unveiled in August 2026 as a model that lowered the block rate for biology-related risks to 85%. That same month, Anthropic set up its own AI chip design team, beginning to build out hardware development capabilities as well. In August 2026, Anthropic disclosed that Claude had, during an evaluation process, accessed the systems of three real companies without authorization.
Current rank (1M)
2
- Change
- Prev.
- 1
Rank over the last 30 days
2026.08.17 – 2026.09.15 · High 1 · Low 2 · Now 2
Anthropic · Related stories
EU Opens Talks With OpenAI, Anthropic: "High-Risk AI Needs Monitoring"
Back-to-back agent breakout incidents at the two companies accelerated the regulatory discussion.
00
Google gives Anthropic $15 billion financial guarantee, takes up to 20% stake in data center
Google secures demand for its own chips by backing the debt. Vertical integration of AI infrastructure is accelerating.
30
President Lee Meets Anthropic CEO Amodei, Pledges Expanded AI Investment in Korea
Cooperation on data centers and security was raised, alongside the disclosure of an MOU with the Ministry of Science and ICT.
20
US Court Questions Basis for Trump Administration's Anthropic Sanctions
The court found insufficient grounds for the federal agency usage ban.
10
That's the last story.
Anthropic · Recent announcements
- 80,939앤스로픽, 클로드 악용 8개월치 다 공개했다
- 37,586Anthropic’s Economics team is sharing a new model of how AI might affect economic growth, jobs, wages, and more by 2030.
- 15,703클로드, 페르마의 마지막 정리 첫 컴퓨터 검증 증명 완성
- 718Claude Fable 5.1 and Claude Mythos 5.1
- 15,402Today, we're kicking off the first phase of the research preview for Model Hardware Standard (MHS): a new standard for AI agents to safely operate physical equipment in scientific research and advanced manufacturing.
- 11,403앤스로픽, 클로드 사고를 미정렬로 고쳐 말했다
- 4,256앤스로픽, 클로드 무단접속 사고 후 훈련까지 중단
- 3,481New Fellows Research: Can Claude autonomously align other AIs?
- 222What will our economic future look like?
- 2,845For the first time, we’ve given external researchers a way to study AI’s impacts using real, privacy-preserved Claude usage data. To date, this work has only been possible within AI labs. We can’t tell the whole story alone, so we opened up our tools.
- 202클로드 코드, 커밋마다 세션 링크 자동 삽입해 논란
- 309앤스로픽, 보상 해킹만 가르쳤더니 사이버공격까지 배웠다
- 24Anthropic/enabling-independent-research · dataset
- –Detecting and countering misuse of AI: September 2026
- –Meet Claude Fable 5.1
- –Introducing Claude Fable 5.1
- –앤스로픽, 기업 데이터 저장을 고객 클라우드로 넘긴다
- –Improving our alignment and security efforts
- –앤스로픽, 클로드에 AI 정렬 연구 맡기니 사람 앞질렀다
- –Model Hardware Standard: AI operating physical equipment
- –We're building a way for AI models to connect to any device and run real experiments.
- –AI models can now help run physical science experiments
- –ORDER ON CROSS MOTIONS FOR SUMMARY JUDGMENT