SDSignal Desk

Inside Anthropic’s 93 pages of risk factors, from rogue models to customer concentration

· Updated · Signal Desk newsroom

Analysis of reporting by Ars Technica. How we work

Image: Ars Technica

Anthropic’s S-1 warns of models that could resist human control, questions of AI moral status and research automation outrunning testing. The grounded risks matter more for now: unsettled agent liability, top-five customers near a quarter of revenue, and rivals that double as compute suppliers.

Key takeaways

  • The S-1’s Risk Factors section runs 93 pages and more than 70,000 words.
  • It flags models that could resist human control, AI moral status and AI research outpacing safety testing.
  • The filing says the most serious autonomy concerns apply to near-future models, not current systems.
  • Customers may sue over agent actions even when they supervised the agent, and contract protections may not hold.
  • The top five customers were nearly a quarter of revenue in the first nine months of 2026.

A risk section like no other

Anthropic’s S-1, the registration statement filed with the SEC before going public, includes a Risk Factors section that runs 93 pages and more than 70,000 words. Most covers the usual worries: competition, costs, regulation. Tucked in are disclosures no ordinary company would write: models that could act to resist human control, the possibility that society decides AI has moral status, and future systems that could automate AI research faster than the company can test them. The filing notes the most serious autonomy concerns apply to near-future models rather than any current system. It also flags reputational spillover if another company’s AI caused harm at scale or the public blamed AI for job losses regardless of the actual employment impact. Earlier leaks of the prospectus carried similar language (earlier S-1 warnings).

Sincere and convenient at once

There is something surreal about warning that self-improving AI could threaten humanity while noting it could hurt quarterly results, and Ars Technica had fun with that. But risk factors are written by lawyers to protect the company, and if leadership genuinely believes the risk exists, leaving it out would be the bigger problem. A securities filing carries legal weight that a blog post does not. Critics argue dramatic risk talk helps frontier labs argue only a few well-funded players should build the technology. A disclosure can be sincere and strategically convenient; the test is whether actions match the warnings, such as Anthropic’s proposed safety bill with explicit harm thresholds (Anthropic’s safety bill).

The risks that will show up first

The grounded risks are more pressing. Anthropic says customers may sue over actions taken by AI agents even when the customer set up and supervised the agent, and it is unclear whether contract protections would hold. It flags liability when models interact with people at risk of self-harm and when minors get around age limits, and for agents operating physical equipment it describes harms that are immediate and irreversible. Real-world damage is likelier to come from an agent making a costly mistake with nobody sure who is responsible than from a sentient Claude.

Then there is business fragility. The top five customers made up nearly a quarter of revenue for the first nine months of 2026, many are not on long-term contracts and many run on venture money. Amazon, Google and Microsoft are key compute suppliers and direct rivals (Amazon’s stake). And the company has committed $125 billion to data centers that could age badly if hardware or training methods shift. Those are the risks that move a stock. As a public record, the filing sets a healthy precedent of disclosing behavioral risk, provided disclosure is not a substitute for reducing it. More in our Anthropic coverage and AI coverage.

Who feels it

Investors
Customer concentration, rival suppliers and big infrastructure commitments are the concrete risks.
Enterprises deploying agents
Liability for agent actions is unsettled; review contracts and supervision.
Regulators
A lab has formally described catastrophic and manipulation risks in a legal filing.
Parents and vulnerable users
Anthropic names minors and people at risk of self-harm as exposed groups.

What to watch

  1. Whether other AI companies adopt similar behavioral risk disclosures.
  2. Early court cases on liability for agent harms.
  3. Changes in Anthropic’s top-customer mix or new long-term contracts.
  4. Progress on direct data center leases versus rivals’ clouds.

FAQ

How long is Anthropic’s risk factors section?

93 pages and more than 70,000 words.

Does Anthropic say current Claude models could resist control?

The filing says the most serious autonomy concerns apply to near-future models rather than current systems.

What business risks does the filing flag?

Top five customers near a quarter of revenue, dependence on rival cloud providers and $125 billion in data center commitments.

Read the original

Continue at the source.

Ars Technica

Companies: Anthropic

More on Anthropic

All Anthropic coverage →