Science & Tech · AI · 1 hr ago
Anthropic gives outside groups access to assess AI safety risks
Anthropic is bringing independent groups into its safety reviews as it assesses risks from its AI systems.
The company is working with nonprofits METR and SecureBio, and is discussing possible work with other organizations, including Hugging Face.
METR has employee-like access to Anthropic's hardware, offices and tools, and will examine risks, incidents and practices before publishing its findings.
SecureBio has more limited access and is reviewing the chemical and biological weapons section of a risk report Anthropic released in August.
The plan follows CEO Dario Amodei's proposal to slow AI model development and his pledge to work with outside evaluators.
Independent checks are gaining importance as concerns grow about whether AI companies can keep their models safe.
Anthropic says it is working toward an expanded partnership with Accenture, while SecureBio could take part in projects with broader access and publication rights in future.
Anthropic has given six researchers from nonprofit METR employee-level access to assess its AI risks and safety practices.
METR can examine Anthropic’s risk levels, incidents and practices, and publish its findings without the company’s approval.
Anthropic may request redactions for information in narrowly defined categories, including customer data and model architecture information.
Nonprofit SecureBio is conducting a more limited review of the chemical and biological weapons section of an Anthropic risk report.
The company is also discussing possible work with other organizations, including Hugging Face.
- Who
- Anthropic and the nonprofit evaluators METR and SecureBio.
- What
- Anthropic is giving METR researchers employee-level access to assess its models, safety practices and risks.
- When
- Anthropic announced the arrangement on October 15, 2026.
- Where
- At Anthropic’s facilities and through access to its internal tools and systems.
- Why
- To have outside evaluators assess its risks and practices and improve public accountability.
This story does not have two clearly opposing sides.
CEO Dario Amodei proposed embedding outside evaluators inside AI labs in an essay.
METR published an evaluation proposal for frontier AI companies.
Anthropic announced employee-level access for six METR researchers and a limited review by SecureBio.
- METR team
- 6 researchers
- METR access level
- Top tier of Anthropic’s three-tier evaluator system
- SecureBio review
- Chemical and biological weapons section of an Anthropic risk report released in August
- Publication rights
- Anthropic cannot veto METR’s findings but can request specified redactions
- Potential partners
- Hugging Face and other organizations are in early-stage discussions
5 sources
Anthropic、METRの評価チームを社内に常駐させ“従業員並み”のアクセス付与 公表に拒否権なし
Anthropic taps nonprofits METR and SecureBio to help assess risks and safety practices
Anthropic is letting METR researchers look under the hood with employee-level access
Anthropic grants non-profit METR ‘employee-like’ access to monitor AI risks
Anthropic just opened its doors to outside safety watchdogs. Here's how it will work.
What people are saying
We want to be publicly accountable for the decisions we’re making
We're embedding evaluators voluntarily, and we believe every frontier AI company should do the same
Anthropic can’t veto publication, but we can require them to redact information if it falls into narrow, pre-defined categories