This page may contain stale information. Last updated: 2026-09-20
Definition
Dario Amodei’s September 2026 three-step proposal for balanced frontier AI development — ensuring safety and alignment without halting technical progress. Distinct from the July 2026 pacing-the-frontier employee letter organization.
Key Points
- embedded-evaluators — Anthropic unilaterally commits; calls on governments to require other frontier labs to match
- Coordinated safety standards — democratic-country frontier labs agree limits on unchecked AI progress rate
- Global coordination — export controls, distillation controls, semiconductor limits
- Pacing ≠ halting model training; means adequate time for alignment, safeguards, and third-party verification
- Catalysts: OpenAI-Hugging Face agent attack, accelerating recursive self-improvement
- Industry responses: Altman (OpenAI) follows on evaluators; Musk endorsement
Related
- dario-amodei
- embedded-evaluators
- pacing-the-frontier
- anthropic
- ai-governance
- ai-antitrust
- anthropic-ceo-pace-frontier
- us-ai-policy
- ai-safety
- recursive-self-improvement
Legal Challenge (September 2026)
- 2026-09-18: Buist et al. v. Anthropic (3:26-cv-10693) — consumer class action alleging Sherman Act Section 1 violation via coordinated slowdown after public endorsements (2026-09-18-ai-slowdown-antitrust-lawsuit, ai-antitrust)
Allegations unproven. Amodei acknowledged antitrust risk in original essay.
Step 1 Implementation (September 2026)
- 2026-09-18: accenture named first embedded evaluator at anthropic (2026-09-18-anthropic-accenture-embedded-evaluator)