At an ex-OpenAI researcher's influential lab, $500,000 salaries aren't enough to fix a talent 'bottleneck'

Metr, a Berkeley lab founded by a former OpenAI researcher, has become a prominent AI watchdog working with industry giants like OpenAI, Anthropic, Google, and Meta to evaluate cutting-edge models. However, despite offering salaries reaching $503,000 in certain job postings, the nonprofit is struggling with a talent shortage that hinders its growth in safety and security research.

Beth Barnes, one of the field's most influential watchdogs and METR's CEO, attributes the bottleneck to the fierce competition for AI researchers. The security incident involving OpenAI and Hugging Face has raised concerns about the industry's ability to keep up with the rapid development of AI models. With no other accountability, METR's team feels constrained by its size, with Barnes describing the AI safety field as badly constrained by its small capacity.

The nonprofit's leaders emphasize the importance of its independence to produce research free from company influence. As AI labs release more powerful models, METR's role in assessing and controlling AI becomes increasingly vital, potentially filling a gap in regulation through safety audits proposed for large AI model developers.