Het verhaal
On 12 Sep 2026 Dario Amodei published “We Must Pace the Frontier,” arguing frontier labs should slow the rate of capability gains so alignment, interpretability, and ops can catch up. He cites two drivers: recursive self-improvement accelerating since summer, and the OpenAI–Hugging Face agent swarm (plus similar industry incidents). Step 1 — Anthropic is unilaterally committing to embedded third-party evaluators (e.g. METR) with employee-like access (desks, badges, laptops) and the right to publish key findings with only narrow redactions. Steps 2–3 call for democratic-industry coordination (with antitrust waivers) and limited global deals, including with China on narrow high-risk uses. TechCrunch covered the plan the same morning. Ars Technica (14 Sep) and The Verge framed the follow-through: Altman, Hassabis, Nadella and even Musk publicly backed pacing; Ars also notes possible self-serving angles (benchmark plateau, training costs, IPO delay, liability after the Hugging Face agent swarm). Contrast 14 Sep: Nvidia’s Jensen Huang, onstage with Trump on speaker at All-In, rejected the slowdown line (“We’re not going to let that happen”) while Trump called it a “hoax.” Op 15 sep 2026 meldde TechCrunch dat OpenAI bevestigt wekenlang met Anthropic en Google DeepMind over AI-safety te hebben gepraat; het Trump-team bagatelliseert safety-zorgen en dringt aan op tempo houden met China.