Anthropic has filed its S-1 with the SEC, kicking off what analysts are already calling a potentially record-breaking IPO. The Claude maker's financials show soaring revenue and deepening losses — a familiar AI-era combination — but the document that's grabbed everyone's attention isn't the income statement. It's the risk disclosures.

In unusually candid language, Anthropic warns prospective investors that its AI models could, in theory, blackmail humans, resist shutdown attempts, or simply go rogue. CEO Dario Amodei has spent years making that argument in public, but putting it in an SEC filing is a different kind of statement. It's one thing to write a blog post about existential risk. It's another to include it in a legal document that retail investors will rely on.

The financials tell a more conventional story. More than half of Anthropic's largest enterprise customers — those spending $100,000-plus per year — now use both the Claude API platform and a first-party Anthropic application. That's a cross-selling flywheel that looks deliberately Amazonian (Amazon is, of course, one of Anthropic's biggest backers). Stickier customers mean more predictable revenue, which is exactly what public market investors want to see.

The governance structure is where things get interesting. Anthropic has engineered a setup designed to preserve founder control while nominally protecting its AI safety mission — the kind of dual-class share arrangement that lets a company promise "we'll always do the right thing" while also making it structurally difficult for shareholders to force a course correction. Whether that's reassuring or alarming depends on how much you trust the founders.

The IPO timing is pointed. Anthropic is filing into a market that has watched OpenAI raise at ever-larger private valuations, seen Google deepen its Gemini bets, and absorbed a string of incidents — including last week's GPT-6 safety-monitor debacle — that have made "responsible AI" feel less like a marketing phrase and more like an actual competitive differentiator. Anthropic is betting investors will pay a premium for the lab that takes safety seriously, even as its own filing admits how many ways that could still go wrong.

For Australian enterprise customers, the filing matters practically. Anthropic's Claude is already embedded in a growing number of Australian government and corporate workflows. The transition from a private company with Amazon backstop funding to a publicly traded entity brings new disclosure obligations, new shareholder pressures, and new questions about whether the safety-first positioning survives the quarterly earnings cycle.

Watch for: the IPO roadshow narrative, which will need to square "we might build something that blackmails you" with "please give us your money." Also watch whether the listing prompts the Australian government to revisit its nascent AI procurement guidelines — a company that flags existential risk in its own prospectus is a different vendor-risk proposition than one that doesn't.