Anthropic Defends Model Oversight as FTC Examines AI Safety Practices
publishers 114articles 195first reported 24 Sep, 16:43 UTCdeveloping since 24 Sep · 38 editions
Anthropic said it intends to retain oversight of how customers use its most powerful AI models, arguing that their “dual use” nature requires safeguards capable of detecting misuse.
Speaking at the opening of Anthropic’s Singapore office, co-founder Benjamin Mann said the company wants customers to run sensitive workloads on its systems while preserving its ability to identify harmful activity. Anthropic’s Enterprise Frontier Safeguards program offers eligible customers protections including zero data retention while maintaining misuse-detection measures.
The comments came as the U.S. Federal Trade Commission sent civil investigative demands to Anthropic and OpenAI. The agency is seeking information about safety incidents, privacy policies, security measures and potentially misleading marketing, according to reports by Axios cited by CBS News and Ars Technica. The FTC confirmed the subpoenas but offered no further details. Additional demands may be sent to third-party organizations.
Separately, Anthropic said it granted six employees of nonprofit evaluator METR access broadly comparable to that of its internal risk-assessment teams. METR may independently publish assessments of Anthropic’s models, incidents and safety practices without prior company approval, subject to limited redactions for customer data, privileged information and some model-architecture details.
Anthropic is also expanding its enterprise operations. The company plans to spend $100 million to train 10,000 forward-deployed engineers by the end of 2027, CNBC reported. Its new Singapore location is its fifth office in the Asia-Pacific region.
HOW THIS STORY WAS MADE
Written from 8 articles, 7 with the publisher's own text; 8 independent newsrooms once syndicated copies count as one; this version written 382 h after the record first saw the story; the editor kept it as written.
- Publishers
- 114
- Source articles
- 195
- Given to the writer
- 8, 7 with the article's own text
- Independent newsrooms
- 8
- This version written
- 382 h after the record first saw the story
- Second model (editor)
- kept as written
- Publication gate
- passed
- Human review
- none
WHAT THE EDITOR DID
The second model checked the writer's draft against the same sources and published it as written.
Written by a language model from the sources above, then checked by a second model that may only cut, attribute or correct. How it works →