Anthropic Lets Outside Watchdog Inside to Check AI Safety

Anthropic, the company behind the Claude AI chatbot, has given an outside safety group called METR special access to its offices and computer systems. METR employees now work alongside Anthropic staff, with the same access as internal risk teams. They can look at how the AI is built and test it for dangers. This is part of a promise to be more open about AI risks.

Why should you care? As AI gets more powerful, it could cause problems like helping hackers or spreading misinformation. Having independent eyes on these systems could catch issues early, making the AI you use safer. It's like having a food inspector in a restaurant kitchen—you don't see them, but they help keep your meal safe.

But there's a catch. Critics say these arrangements can lead to conflicts of interest, like when accounting firms got too cozy with the companies they audited. Also, Anthropic can still keep some secrets, like how the AI works under the hood. And METR isn't paid by Anthropic, which helps independence but raises questions about funding.

This move puts pressure on other AI companies like OpenAI to follow suit. If it works, it could become a model for how we keep AI in check. For now, it's a first step. The real test is whether METR's reports lead to real changes—and whether you'll notice a difference in how safe AI feels.