Anthropic IPO Could Raise $64B—Plus 4 AI Stories
Anthropic’s IPO filings show 13x revenue growth, while DeepMind proves Signal’s security in weeks.
Anthropic’s IPO files show a company growing fast
New York is forcing big AI labs to register
DeepMind’s Signal proof is a milestone—with limits
ChatGPT opens up, and Upwork writers feel the squeeze
## Anthropic’s IPO files show a company growing fast
Anthropic’s public filing puts a price range of $70 to $78 on the 825 million shares it plans to sell, with the top of that range raising $64.35 billion before bank fees. For developers, the headline number is less important than the shape of the business: revenue for the first nine months of 2026 is $33.24 billion, about 13 times the year-earlier figure, while operating losses narrowed from $5.22 billion to $1.26 billion. The $104.0 billion net loss looks scary until you notice that $94.3 billion is a non-cash accounting charge tied to convertible notes. Pricing is set for 18 November, trading begins on 19 November, and about 20% of shares are reserved for ordinary investors.
## New York is forcing big AI labs to register
Under the RAISE Act, labs with more than $500 million in annual revenue that train models using more than 10^26 operations must register with New York’s financial regulator by 1 January. From then, they have 72 hours to report a serious safety incident. Covered labs must publish a safety plan, file a quarterly catastrophic-risk summary, and issue a transparency report with each major model release. Fines start at $1 million and rise to $3 million for repeat violations. If you build infrastructure or models at that scale, this law is now a compliance requirement, not a proposal. The state is also funding a Binghamton University testing center with more than 80 researchers.
## DeepMind’s Signal proof is a milestone—with limits
A five-person Google DeepMind team used AI-written Lean proofs to verify security properties of libsignal, the library behind Signal’s encryption. The proofs cover about 6.5 million lines of compiled machine code, not a rewritten specification, and the work took four weeks. That is a dramatic speed-up compared with the seL4 kernel, which took roughly twenty person-years to verify. But the proof covers one-to-one sessions only; group chats, server code, and sealed sender are not included, and cryptographic primitives are assumed rather than proved. The team demonstrated the theorem catches regressions by reinserting two old bugs and showing the proof fails. Still, the social media claim goes beyond what the paper states.
## ChatGPT opens up, and Upwork writers feel the squeeze
OpenAI is letting businesses plug their own sales chatbots into ChatGPT, giving those bots access to an audience of 1.2 billion weekly users. For developers, that is a potential distribution channel, but it also raises questions about moderation and liability. Meanwhile, a new study shows freelance writing losing ground on Upwork, a reminder that generative AI is already reshaping the market for text work even as the technology itself advances.
For developers, the most useful takeaway is not the IPO number but New York's registration deadline and DeepMind's proof speed-up.