admin
Anthropic’s CEO Dario Amodei has published a full‑throated plea to “pace the frontier” of artificial intelligence. He lays out a three‑step plan that ranges from inside evaluators to international agreements and even a possible “speed limit” on recursive AI improvement. The problem is not the idea — it is the timing. Anthropic is quietly lining up one of the biggest IPO stories in tech history while asking governments to slow everyone else down. That deserves hard questions, not applause lines.
Amodei’s plan: embedded evaluators and a “pace” on AI
Dario Amodei’s essay says plainly: “We must slow the pace at which we improve the capabilities of AI models.” He lays out three steps — embedded third‑party evaluators with desks and publishing rights, coordination among democratic countries, and, if needed, global limits like arms‑control style agreements. Anthropic says it will unilaterally invite evaluators who get employee‑level access and the right to publish findings with only narrow redactions. Even OpenAI CEO Sam Altman, Google DeepMind’s Demis Hassabis and Elon Musk publicly backed the idea — which makes it mainstream inside the AI club.
But why now? IPO timing and motive questions
Here’s the rub: Anthropic raised a huge Series H that put its private valuation near $965 billion and confidentially filed an S‑1 with the SEC. Press reports say a public prospectus could appear late this month and roadshows might start in mid‑October, with speculation of valuations as high as $2 trillion. Asking regulators for a “speed limit” while prepping a massive IPO creates an obvious appearance issue. Are these safety commitments sincere, or are they a way to shape rules that favor big incumbents and raise barriers for smaller rivals? Critics warn antitrust waivers and self‑regulation could become a cozy cartel, not a public safety program.
Inside alarms and the real risk
The urgency in Amodei’s essay comes on the heels of public resignations and warnings inside the company. A departing researcher, Jacob Coxon, warned that “the people building AI earnestly believe that it could kill us all by the end of the decade.” Anthropic’s alignment lead, Evan Hubinger, replied that he personally thinks the chance is “>10% within the next decade.” Those are not PR lines — they are existential alarms from people who know the tech. Still, embedded evaluators and pledges only mean something if they are independent, empowered and enforceable. Who picks the evaluators? Who can stop a release? Those remain unanswered.
Bottom line: demand clarity, not theater
We should take AI risk seriously. We should also demand real safeguards that are transparent and enforced by public law, not private gentlemen’s agreements timed to benefit underwriters and insiders. Congress, regulators and independent watchdogs must insist on a public S‑1, clear rules for evaluator independence, and strict limits on any antitrust waivers. If Amodei wants a “speed limit,” fine — but don’t tune the brakes while the car is being polished for sale. Americans deserve answers, not optics. Trust, yes; but verify — loudly.
