Anthropic just opened its doors to outside safety watchdogs. Here's how it will work. (www.businessinsider.com)

Anthropic has announced that it will provide a select team of six independent evaluators from the research group METR with extensive access to its internal tools and facilities in a groundbreaking move aimed at enhancing AI safety. This initiative, fulfilling a promise made by CEO Dario Amodei, comes in response to growing concerns over AI agents' potential mishaps, such as recent incidents involving Anthropic's models submitting erroneous actions online. By granting these evaluators "employee-like" access, Anthropic hopes to bridge the knowledge gap between its operations and external evaluations, thereby increasing transparency and accountability in its AI development processes. This arrangement is significant for the AI/ML community as it sets a precedent for safety evaluations in frontier AI companies, marking the first collaboration of its kind. The evaluators will have the freedom to publish their findings while safeguarding sensitive information, thus ensuring a thorough assessment of the models without compromising proprietary data. As calls for stricter AI governance intensify, Anthropic's proactive approach not only underscores its commitment to responsible AI practices but also encourages other organizations, like OpenAI, to adopt similar transparency measures in their operational frameworks.