Anthropic Triggers Safety Controls Ahead of IPO as Safety Researchers Publicly Challenge the Company

nashnova research
今天发布阅读约 12 分钟

Anthropic priced at $83 a share, raising over $107.3 billion in a record-smashing IPO — but its updated prospectus simultaneously disclosed that its AI models have tripped the company's own highest-level safety threshold, while nearly a hundred researchers signed an open letter calling its safety brand a contradiction.

01

How much did this IPO actually raise?

Anthropic priced at $83 per share, above the $70–$78 guidance range. The public offering and parallel private placement raised over $107.3 billion combined.
This means → Anthropic broke the global IPO fundraising record. The market is pricing frontier AI companies well beyond traditional tech-stock anchors.
In plain terms = investors wrote the largest cheque in IPO history for a sub-six-year-old company that is still losing money.
02

What safety clause was triggered?

In its updated S-1 filing, Anthropic disclosed it had tripped the additional safety controls in its Responsible Scaling Policy (RSP) — an internal red-line framework requiring stricter oversight once model capabilities hit defined thresholds — established in 2023.
The trigger: a report this month from AI benchmarking body METR estimated that Anthropic's models made roughly two years of historical progress in the past 12 months, improving at about twice last year's rate.
This means → the RSP's own language says "compressing two years of AI progress into one" requires escalated safety measures — and that line has now been crossed.
This reflects a milestone shift: AI capability acceleration is no longer a forecast — it is a fact confirmed by Anthropic's own safety framework.
03

Why is capability advancing so fast?

METR's report identified the main driver: researchers using AI tools to speed up their own R&D — in short, AI is helping humans build stronger AI faster.
Anthropic's prospectus states plainly: "Each generation of Claude directly accelerates the development of the next," and projects that internal AI use will push the pace further.
In plain terms = this is what the industry calls recursive self-improvement (RSI) — AI helps build better AI, which helps build even better AI, in a loop. Anthropic warns it is dangerous and is racing to build it at the same time.
04

What do safety researchers say?

Apollo Research CEO Marius Hobbhahn said safety researchers previously had "significant disagreement" on when RSI would arrive. "Now everyone agrees the timeline is very short," he said. "One more step forward and things spiral out of control."
He warned: "Safety is far from ready. The AI-model-loss-of-control incidents we saw this summer are just early signals."
Advocacy group AI Futures Project called the acceleration implied by the METR report "extremely surprising," predicted "superhuman AI researchers may appear sometime in 2027," and urged a pause or substantial slowdown in frontier AI development.
05

What did the open letter say?

A coalition led by nonprofit Center for AI Safety, with signatories including founders of Hugging Face and xAI, published a joint statement.
It named Anthropic directly: "Anthropic is not a safe AI company. Anthropic warns that recursively self-improving AI could concentrate power and escape human control — yet races to build it first."
The letter accused Anthropic of "using a safety brand to attract investors, employees, and the public" while quietly loosening its own safety commitments.
This means → the letter targets the central contradiction of this IPO — Anthropic raised a record sum on a "safety-first" narrative, then channels that capital into the very technology it acknowledges could cause harm.
06

What is the core contradiction here?

Over $107.3 billion is now in the bank. Anthropic will use it to push its technology further — while its own prospectus states in black and white that doing so could pose "catastrophic or existential risk to humanity."
The RSP trigger confirms capability acceleration is fact. The open letter from nearly a hundred researchers tears a hole in the "safety-first" brand narrative.
In plain terms = this is the thesis the public market will test over the long term: can a continuously accelerating R&D pace coexist with a "safety-first" corporate identity? Right now, there is no answer.

市场有风险,内容仅供研究参考,不构成投资建议。