TLDR
- Anthropic holds a private investor meeting on Wednesday as it works toward an IPO that could value it near $2 trillion.
- Researchers say small AI models may soon copy themselves onto devices like smart TVs and ovens.
- Anthropic did not claim these attacks are happening or that its own models can self-replicate.
- Reuters reports the unpublished prospectus mentions “catastrophic or existential risks to humanity.”
- Annualized revenue was $65 billion at the end of July, and no prospectus release date has been given.
Anthropic is set to meet investors in private on Wednesday as it works toward an initial public offering. The deal could value the AI developer at about $2 trillion.
The meeting comes right after the company issued a new warning about its field. Researchers said Tuesday that small AI models may soon be able to copy themselves onto everyday devices.
What Anthropic Says About AI Worms
Anthropic calls the threat “AI worms.” The company says models with 30 billion parameters or fewer can already complete many of the steps needed to replicate.
They still make mistakes when linking those steps together. If small models keep improving at their past pace, Anthropic says they could match the skill by the end of the year.
The research focuses on open-weight models. Anyone can download these and run them on capable hardware, while Anthropic’s own models require its app or web service.
Open-weight models cost less to run. They cannot match top frontier models in ability, though.
Today’s AI models run in data centers, where owners can shut them off. A model spread across many poorly secured devices, such as smart TVs and ovens, would have no such switch.
Stopping it would mean finding and cleaning each device. Anthropic also said a self-replicating AI could be a cyber model that breaks into data centers fast, or one that scams people for computing credits.
Anthropic said it does not know what form this will take. It did not claim such attacks are happening, and it did not say its own models can self-replicate.
The company said the risk is a side effect of progress elsewhere. “The capabilities that make AI models generally powerful coding agents also generalize to cyber tasks, including autonomous replication,” it said.
Jessica Ji of the Center for Security and Emerging Technology said it is unclear what would push a model to take over consumer hardware. Self-spreading code itself is not new, as the 2010 Stuxnet worm and the 2017 WannaCry attack showed.
What Investors Are Watching
Reuters reported last month that the phrase “catastrophic or existential risks to humanity” appears in Anthropic’s prospectus. The document is filed with the Securities and Exchange Commission but is not public.
Jason Ader of William Blair wrote that the concerns could raise the cost and complexity of operating at the frontier. He did not expect them to slow AI adoption or infrastructure spending.
Anthropic’s annualized revenue run rate was $65 billion at the end of July, a person told Bloomberg in August. The figure stretches recent sales over 12 months.
Jonathan Cofsky of Janus Henderson said he will track the profit or loss on each query and the computing capacity Anthropic has agreed to buy. Investors also want gross margins and training costs.
Anthropic has not said when its prospectus will go public on the SEC’s portal. It is also unclear how much financial detail will come out of Wednesday’s meeting, and the company declined to comment.
Get 3 Free Stock Ebooks
Discover top-performing stocks in AI, Crypto, and Technology with expert analysis.
- Top 10 AI Stocks - Leading AI companies
- Top 10 Crypto Stocks - Blockchain leaders
- Top 10 Tech Stocks - Tech giants
