EU AI Office adopts AEF-1 standard for independent frontier model evaluations
Miles_Brundage · x · 2026-10-22
The EU AI Office has published requirements for qualified independent evaluations of frontier AI models. In bilateral communication, the Office clarified that evaluator independence under GPAI Code of Practice Measures 3.2/3.5 must be ensured via contractual safeguards incorporating all requirements from AEF-1's Analytic Autonomy and Transparent Methods and Results sections.
AEF-1, from the AI Evaluator Forum, sets minimum operating conditions—independence, access, transparency—across five principles, with evaluators publishing a checklist alongside results. In-practice examples include Transluce's assessment of 77 model variants' mental-health responses and METR's frontier risk report on agents deployed internally at Anthropic, Google, Meta and others.
More from Safety
- Beginner-friendly Mac security audit guide with ready-made prompts for Claude and Codex — AIandDesign · 2026-10-22
- Dev who leaked his own .env via AI coding tool, got a $90,000 Google bill, gets roasted online — weswinder · 2026-10-22
- Toby Ord: Gemini 4 Safety Report discloses DeepMind's internal AI-for-AI-R&D efficiency data — tobyordoxford · 2026-10-22
- The Atlantic essay: build tools to attribute AI output to pre-training works — TuhinChakr · 2026-10-22
- Counterintuitive AI safety take: perfect containment for 10 years may be worse — CedricMakes · 2026-10-22
- Google DeepMind faces resignations over AI military contracts as staff say gag clauses silence them — BlackHC · 2026-10-22