Require AI labs to publish safety evaluations before deploying frontier models
Equally concerning are reports that new LLM models are being sped to market without basic safety testing, or with the results of that testing ignored.AI Verified source (Jan 15, 2026)