Dario Amodei says Anthropic is "unilaterally committing" to giving third-party evaluators permanent access to verify its adherence to safety measures

AI safety audits will become more rigorous and transparent for companies following Anthropic's lead.

Anthropic CEO Dario Amodei announced that the AI company will permanently grant third-party evaluators access to its systems at an employee level. This move is intended to verify Anthropic's compliance with its safety measures. Amodei stated this is the first step in a three-part plan for the AI industry to slow down its rapid advancement. He also referenced previous essays discussing AI policy, national security risks, and the need for industry-wide collaboration, such as Project Glasswing, to address AI-driven cyber threats.

Anthropic's commitment to granting permanent, employee-level access for third-party safety evaluations sets a new precedent for AI development transparency. This initiative directly addresses growing concerns about AI safety and control, potentially influencing regulatory approaches and competitive practices across the industry. Companies may feel pressured to adopt similar measures to maintain public trust and avoid regulatory scrutiny.

The move signals a potential shift towards proactive safety verification rather than reactive policy-making in the AI sector. By allowing deep access, Anthropic aims to build confidence that its advanced models are developed and deployed responsibly. This could accelerate the development of standardized safety assessment frameworks and tools, making AI development more accountable to external oversight.