Anthropic's IPO warning and AI's credibility problem

Anthropic, the AI company founded by former OpenAI researchers, has filed for an initial public offering that BBC News describes as “blockbuster” — while simultaneously publishing disclosures warning of “catastrophic risk to humankind” from the technology it is rushing to market. The IPO, detailed further by BBC analysis, could value the company in the hundreds of billions and would rank among the largest technology listings in history. NPR reports that the filing could break records for the sector. The company, which markets itself as the “safety-focused” AI lab and has positioned its Claude models as more responsible alternatives to competitors, is seeking to raise capital from the same public markets it has warned may be funding an existential threat.

The received wisdom

The AI safety community and much of the technology press have largely accepted Anthropic’s framing at face value. On this reading, the company occupies a difficult but honourable position: it believes powerful AI is coming regardless of what any single actor does, and therefore it is better to have safety-conscious labs at the frontier than to cede that ground to less scrupulous competitors. The “race to the top” argument holds that Anthropic’s commercial success enables the safety research that might prevent the catastrophes its risk disclosures describe. IPO proceeds fund interpretability research, red-teaming, and policy engagement. The alternative — staying private and falling behind on compute — merely hands the frontier to OpenAI or Chinese competitors with less robust safety cultures. Within this frame, the apparent contradiction between catastrophic risk warnings and aggressive fundraising is not hypocrisy but a calculated wager on a second-best outcome.

A different read

There is something to that argument. But it deserves more sceptical scrutiny than it typically receives, and the IPO filing is a useful moment to apply it.

The first problem is incentive alignment. Once Anthropic has public shareholders, it has a legally enforceable duty to maximise returns for those shareholders. The history of technology companies that began with strong ethical missions and then had to manage the tension between those missions and the demands of public markets is not encouraging. Facebook launched with a “make the world more open and connected” mission statement; the fiduciary duties of a public company, combined with advertising-based revenue models, produced outcomes that differed somewhat from that vision. Anthropic’s safety commitments are currently enforced by the personal ethics of its founders and the reputational incentives of a private company in a narrow industry. Post-IPO, those enforcement mechanisms will be supplemented — and in time may be overwhelmed — by quarterly earnings pressure.

The second problem is the specific content of the risk disclosures. BBC reporting notes that Anthropic itself warns the technology could pose “catastrophic risk to humankind.” This language appears in documents filed with regulators as a legal requirement of material disclosure. The company is therefore simultaneously telling the Securities and Exchange Commission that its product might end civilisation and telling investors that this is a good time to buy. If the catastrophic risk language is genuine, then the IPO is deeply problematic on ethical grounds. If the language is boilerplate designed to manage legal liability, then the company’s much-publicised safety commitments are subject to reinterpretation in the same direction.

Third, there is the question of what “safety-focused” actually means in practice. The UK invested £100m in an AI Biosecurity Lab partly in response to concerns raised by companies including Anthropic. Government policy is therefore being shaped by the same actors who stand to benefit financially from the regulatory frameworks that policy produces. This is regulatory capture in its most sophisticated modern form: not the old-fashioned variety where an industry association lobbies for looser rules, but the new variety where industry leaders advocate for specific safety regulations they are best positioned to comply with, thereby raising barriers to entry for competitors.

None of this means AI safety concerns are not real — they may be very real indeed. But the vessel carrying those concerns is a company filing for a record IPO. The market is entitled to scrutinise the cargo more carefully.

What to watch

  • IPO pricing and lock-up periods: how Anthropic’s founders and early backers structure their equity positions post-listing will reveal how seriously they take their own catastrophic risk language.
  • Regulatory response: whether US, EU, or UK regulators use the IPO filing’s risk disclosures as the basis for mandatory safety requirements — or simply file them away as legal boilerplate.
  • Safety research funding proportion: post-IPO annual reports will show what share of revenue actually goes to safety research versus product development. The ratio will be revealing.
  • Competitor reactions: OpenAI and Google DeepMind will be watching closely; if Anthropic’s safety positioning earns a valuation premium, expect rapid mimicry across the sector.

— J