Hugging Face Policy Chief Flags AI Evaluator Limits
Original: Embedded AI evaluators won’t be an easy solution, Hugging Face’s policy chief says
Why This Matters
Evaluation credibility underpins AI safety claims and regulatory trust industry-wide.
Irene Solaiman, Hugging Face's chief policy officer, warned at TechCrunch Disrupt that embedded AI evaluators lack standardization, reproducibility remains unsolved, and current benchmarking methods are immature.
At TechCrunch Disrupt, Hugging Face's chief policy officer Irene Solaiman pushed back on growing industry enthusiasm for embedded AI evaluators — third-party observers placed inside AI organizations to assess model safety and performance — calling the space far from ready.
Solaiman pointed to two core problems: reproducibility and comparability. "Can you reproduce the results of an evaluation? Do you have the level of access that you need to the model itself — this is where open weights come in in a big way," she said. Benchmarks also become outdated quickly, and there's no agreed method for comparing results across different evaluations.
The push for embedded evaluators has grown as AI labs face mounting safety incidents — Hugging Face itself was a notable target. Rather than accept government regulation, labs have promoted the third-party model. But Solaiman cautioned that "the qualifications that determine who is an effective evaluator across what tasks need a lot more standardization." Anthropic's decision to hire Accenture as its first evaluator drew sharp criticism from researchers who questioned whether a consulting firm was equipped for the job.
Solaiman also highlighted a shift in what evaluations must cover: it's no longer just the model, but the full system — including the harness (the tool users employ to access the model). A weaker model paired with a stronger harness can outperform a frontier model, she noted, adding further complexity to any standardization effort.