Anthropic taps nonprofits METR and SecureBio to help assess risks and safety practices
The company is also engaged in early-stage discussions with other organizations, including Hugging Face.
TL;DR
- Anthropic is collaborating with nonprofits METR and SecureBio for independent safety and risk assessments.
- Discussions are underway with other organizations, including Hugging Face, for potential evaluation partnerships.
- The AI lab is implementing a tiered system for evaluators, offering different levels of access and autonomy.
- METR, a third-tier evaluator, has employee-like access to Anthropic's resources to analyze risks and practices.
- SecureBio, a first-tier evaluator, has more limited access and is reviewing specific risk report sections.
- The need for independent AI evaluation has grown amid increasing concerns about AI model safety.
- Anthropic acknowledges the current lack of standards and settled funding systems for independent evaluations.