AI Safety Evaluators Accused of Conflicts of Interest, Staff Turnover Sparks Calls for Congressional Investigation

AI Safety Evaluators Accused of Conflicts of Interest, Staff Turnover Sparks Calls for Congressional Investigation

An analysis that garnered nearly 5 million views on social media has thrust the financial ties between AI safety evaluators and frontier labs into the spotlight. At the heart of the controversy are the frequent personnel moves between evaluators and the companies they assess, as well as potential conflicts of interest in the funding chain.

Take METR (Model Evaluation and Threat Research), one of the main organizations assessing the deployment safety of frontier AI models. Its staff includes individuals who previously worked at OpenAI and Anthropic. On the funding side, Facebook co-founder Dustin Moskovitz holds equity in Anthropic, and his philanthropic foundation has channeled substantial funds through effective altruism networks to evaluators such as METR and Redwood Research. This means that money flowing to safety watchdogs can sometimes be traced back to the companies being evaluated in just one or two steps.

Moreover, some evaluators have personal or family ties to employees at the labs they assess. Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman have previously proposed embedding third-party evaluators directly inside labs, further heightening concerns. Calls are now growing for Congress to investigate.

Source: https://cryptobriefing.com/ai-safety-watchdog-conflict-of-interest/