Anthropic grants external AI watchdog 'employee-like' access

Anthropic has made a significant decision by granting external evaluators 'employee-like access' to assess the risks associated with their artificial intelligence models. This measure is in response to a commitment by the company to enhance safeguards amid growing concerns about the potential for AI technology to spiral out of control.

The organization METR, a nonprofit that has collaborated with leading labs like OpenAI and Google, now has a team embedded in Anthropic's offices. This team will have access to the company's hardware and internal systems, enabling a more thorough evaluation of the AI models developed by Anthropic.

Why it matters

Involving external evaluators in the AI development and assessment process is a crucial step in ensuring that these technologies are managed safely and responsibly. As AI systems become more advanced, the risk of potential harm increases, making such oversight essential for public and industry trust.

What we know

Anthropic's initiative is part of a broader approach by technology companies to address safety concerns regarding AI. Increasing the role of external evaluators is one of the concrete actions proposed to mitigate risks associated with more capable AI systems.

What remains unclear

It is still unclear how the evaluation process will be conducted and what specific metrics will be used to determine the risks associated with Anthropic's AI models. Additionally, it remains to be seen how this collaboration will influence the future development of their technologies.