🔒Anthropic Grants External Evaluators Employee Access to Assess Risks
Third-party evaluators get unprecedented access to Anthropic's systems
TL;DR
Anthropic has given METR evaluators full access to its systems and hardware, a move aimed at reassuring investors and addressing safety concerns. Critics argue this could lead to conflicts of interest.
Anthropic has granted external evaluators, specifically METR, 'employee-like access' to assess risks associated with its AI models. This move is a direct response to rising concerns about the dangers of more capable AI systems. For developers and teams working with AI, this signals a shift towards more transparent and accountable practices in the industry. METR's team of six evaluators will have access to Anthropic's hardware, internal systems, and tools, similar to how bank regulators work inside major banks. This arrangement is set to begin without any financial compensation from Anthropic, and evaluators are expected to publish their findings without Anthropic's veto power.
Key Points
Anthropic has onboarded six METR employees to assess AI risks, giving them access to company hardware and internal systems.
Evaluators will operate like bank regulators, assigned to work inside the largest banks to assess risks.
METR evaluators will publish their analysis without Anthropic's ability to veto publication.
Anthropic is still working with Accenture for safety testing, named a partner last month.
Another third-party tester, SecureBio, reviews chemical and biological weapon risks for Anthropic.
Why It Matters
If you're developing AI models, Anthropic's move to embed external evaluators signals a new era of transparency and accountability. This could set a precedent for other AI companies, impacting how they handle safety and risk assessments. However, critics argue this could lead to conflicts of interest, complicating the landscape for developers and investors alike.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,567 builders reading daily.