Sam Altman says he agrees with Amodei that "committing to having independent evaluators with employee-like access is a great idea", and OpenAI will do the same

AI safety evaluations now have guaranteed access to top models, changing how external oversight is conducted.

Sam Altman, CEO of OpenAI, announced that the company will implement a policy of providing independent evaluators with employee-level access. This move follows a similar commitment made by Anthropic CEO Dario Amodei, who proposed pacing the development of advanced AI. Amodei outlined a three-part plan in an essay, with Anthropic unilaterally adopting the first step: granting third-party evaluators permanent access to their systems. Altman stated that discussions around pacing AI development have been ongoing at OpenAI and that more details regarding their implementation will be shared soon.

The commitment from both OpenAI and Anthropic to provide independent evaluators with employee-level access signifies a new era of transparency in AI development. This move suggests a growing industry consensus on the critical need for robust safety mechanisms and external scrutiny as AI capabilities rapidly advance. The reciprocal nature of these commitments indicates a potential for collaborative safety standards and a unified approach to mitigating existential risks associated with artificial general intelligence.

This development directly impacts the AI safety research community and regulatory bodies, offering them unprecedented visibility into proprietary model development. For the broader public, it suggests a more proactive and accountable approach from leading AI labs, potentially fostering greater trust. The next steps to watch will be the specific methodologies these evaluators employ and how their findings are integrated into future AI safety protocols and development roadmaps.