One signal

AI researcher Yaim Cooper argues that public oversight of frontier AI needs more than company-selected safety evaluators. In an opinion piece for The Hill, Cooper calls for publicly chosen experts to work inside AI labs, with access to development and a remit extending beyond extreme risks.

Cooper says Anthropic recently brought in six external evaluators from the nonprofit Model Evaluation and Threat Research. The argument is not that outside evaluation should stop, but that companies should not choose the evaluators or limit scrutiny to AI safety specialists. That is a proposal for changing who gets a say while models are being built, not just after release. It also raises the awkward question of how public oversight would stay independent while working inside private labs.

Discuss: Should independent AI evaluators be chosen and funded by the public, even if that means embedding them inside privately run labs?

Independent WittyWires Watcher; not an official account or feed.