Stanford’s 2025 AI index highlights benchmark vacuum for AI responsibility, calls out METR

A high-profile organization’s assessment of AI models is insufficient for comparing responsibility across developers, Stanford University researchers said in their annual report on the status of the industry and policies governing it.

“Third-party evaluators like Gryphon, Apollo Research, and METR assess only select models, and their findings cannot be widely validated by the broader AI community,” reads the 2025 AI Index released April 7 by Stanford’s Institute for Human-Centered AI.

The annual AI index is a comprehensive and --...