AI Companies’ Shaky New Plan to Police Themselves
Third-party evaluators will be allowed inside companies to monitor the models. But how much access, and influence, will they really have?
AI companies, including Anthropic, OpenAI, and xAI, have agreed on a new plan to police themselves by introducing "embedded evaluators" to monitor safety practices. These evaluators will consist of outside experts from small nonprofits and large firms, such as Accenture and the nonprofit METR. The evaluators will have access to the companies' internal operations and data, with the aim of identifying risks and writing reports detailing their findings.
However, the details of the system remain vague, with questions about how it will work and the extent of the evaluators' independence. While embedded evaluators are preferable to the current lack of oversight, concerns remain about the potential conflicts of interest and lack of true independent oversight.
Written by urgent.news from The Atlantic's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
