More than 100 AI researchers, including Geoffrey Hinton, signed a public letter establishing five minimum requirements for third party organizations embedded within frontier AI labs to assess model risks. The demands include granting evaluators access equivalent to highly privileged employees and protecting them from retaliatory litigation for findings that are unflattering to the companies.

Signatories require that these evaluators maintain full editorial control, avoid any payment contingent on findings, and disclose conflicts of interest. The initiative is led by the AI Evaluator Forum, which is launching an open call for members and funding to implement the AEF 1 standard after frontier AI companies proposed embedding external oversight to monitor escalating capabilities.