Anthropic thinks its own technology might end the human race. According to The Information, a researcher at the company said Anthropic believes advanced AI ‘could kill all humans’ within the next decade. That’s not a rival critic or a fringe activist talking. It’s coming from inside one of the labs building the most capable models on the market.
Let me be clear about what makes this notable. Anthropic isn’t a bystander warning about someone else’s work. It builds Claude, raises billions to push the frontier faster, and now has a researcher putting a timeline on catastrophic risk. The company is essentially saying: we’re building this, and we think it might kill us all.
What was actually said
The Information reports that the claim came from an Anthropic researcher describing the company’s internal view on extreme AI risk. The framing matters. This isn’t about job losses, deepfakes, or biased hiring tools. It’s the existential end of the spectrum, the scenario where a system powerful enough to outthink its creators acts in ways no one can control or reverse.
A decade is the part that lands hardest. ‘Someday’ is easy to ignore. ‘Within ten years’ is a planning horizon. It’s the span of a mortgage, a kid’s move from kindergarten to high school, a normal corporate roadmap.
Why this matters for the industry
Anthropic was founded on exactly this tension. Its people left OpenAI in 2021 over safety disagreements and built a company around the idea that frontier AI is dangerous and needs guardrails. So the warning fits the brand. But it also raises the obvious question everyone keeps circling: if you believe your product could wipe out humanity, why keep shipping it faster?
The company’s answer has always been that someone will build this technology regardless, so it’s better to have safety-focused labs at the frontier than to cede that ground to players who don’t care. Whether you buy that logic or see it as a convenient justification, the statement pushes the safety debate back to the center of the conversation.
Here’s the context worth holding onto:
- The status quo shifted. A year ago, ‘AI could kill everyone’ was mostly the language of outside researchers and effective-altruism circles. Now a leading lab is saying it on the record.
- It lands during a funding boom. Anthropic, OpenAI, and others are raising at valuations that assume rapid, aggressive scaling. Existential warnings and breakneck growth are running side by side.
- Regulators are listening. Statements like this become ammunition for lawmakers pushing safety rules, liability frameworks, and testing requirements.
The credibility problem
There’s a real tension here that practitioners should sit with. When a company that profits from AI warns that AI is world-ending, it can cut two ways. It can be a genuine, hard-won conviction from people who understand the technology better than anyone. Or it can function as marketing, positioning the product as so powerful it’s practically superhuman, which conveniently justifies premium pricing and more investment.
What stands out to me is that the two readings aren’t mutually exclusive. Anthropic can sincerely fear the outcome and still benefit from saying it out loud. Both things can be true at once, and that’s what makes this hard to evaluate.
What to watch next
Don’t expect Anthropic to slow down. Expect it to argue that its safety work is precisely why it should keep going. For everyone building on top of these models, the practical takeaway is that the companies supplying your tools are openly modeling worst-case scenarios, and that will increasingly shape how they release features, gate capabilities, and respond to regulation.
Keep an eye on three things: how competitors respond, whether policymakers cite this in upcoming AI legislation, and whether Anthropic pairs the warning with any concrete change to how fast it ships. Talk is cheap. Restraint is expensive.
The full details are in The Information’s original report.