Anthropic Grants External AI Watchdog Employee-Like Access

AI company Anthropic has granted an external evaluative organization "employee-like access" to its internal systems and hardware, aiming to increase safeguards amidst concerns about the technology's potential to develop uncontrollably.

METR, a non-profit organization that has previously collaborated with leading AI labs such as OpenAI, Google, and Meta, has now embedded a team within Anthropic's facilities.

A greater role for third-party evaluators is among the key concrete steps proposed by AI labs in response to rising concerns about the potential dangers of more capable AI systems to the economy, cybersecurity, and humanity at large.

This access is intended to facilitate a more comprehensive and in-depth risk assessment, with the goal of anticipating and mitigating potential negative consequences as AI technology advances.