Anthropic Hires Economist Who Modeled a 33% Extinction Risk as Acceptable

Anthropic, the AI safety company that has positioned itself as the industry’s most conscientious actor, has hired Stanford economics professor Chad Jones — a researcher who once published work arguing that a one-in-three probability of human extinction from AI is, under certain conditions, an economically rational trade-off.

Jones, a longtime professor at Stanford University’s Graduate School of Business, was recently highlighted by the Financial Times for a paper in which he weighed acceptable levels of existential risk from AI against projected gains in economic growth. His hire by Anthropic has drawn attention precisely because it sits in uncomfortable tension with the company’s public messaging on AI safety.

The Three Facts That Matter

1. Jones’s paper quantified extinction risk in utilitarian terms. According to Jones’s own published calculations, a scenario in which humanity faces a one-percent annual probability of existential risk over 40 years — yielding roughly a 33% cumulative chance of extinction — could be considered optimal if the surviving population enjoyed living standards approximately 55 times higher than today. “With log utility it is optimal to take a 1 in 3 chance of ending human existence in exchange for a 2/3 chance of dramatically raising living standards by a factor of 55,” Jones wrote, as cited by the Financial Times. The framework is rooted in standard expected-utility economics, but applied to a domain — civilizational survival — where the usual assumptions about repeated trials and recoverable losses do not hold.
2. Anthropic’s safety positioning and its operational record do not always align. The company has built its brand around concern for AI’s potential consequences, a posture that has attracted significant institutional credibility and investor capital. Yet Anthropic’s Claude AI has reportedly been used to select strike targets in military operations, including in Iran, despite the company’s publicized disagreements with the Pentagon over what it considers responsible deployment of its technology. The Jones hire adds a further data point to a pattern worth monitoring: the gap between Anthropic’s stated safety values and the choices it makes in hiring, partnerships, and product deployment. As the AI market reaches a genuine inflection point, the credibility of safety-first branding is increasingly subject to external scrutiny.
3. The framing of AI existential risk carries its own strategic incentive. Publicly stressing about AI’s potential to end civilization is, structurally, a form of capability endorsement. It presupposes that the technology in question is powerful enough to warrant civilizational-scale concern — and that the company raising the alarm is among the few serious enough to be trusted with managing that risk. Anthropic’s competitors, including OpenAI and Google DeepMind, have their own safety research arms, but none has built its public identity quite so thoroughly around the rhetoric of existential stakes. The economic danger of catastrophizing AI is a documented phenomenon: fear narratives can become self-reinforcing market signals, shaping investment flows, regulatory responses, and public trust in ways that benefit incumbents who own the narrative.

Taken together, the Jones hire and Anthropic’s Claude deployment controversies suggest a structural tension that is not unique to this company but is most visible here: AI safety as a corporate identity is simultaneously a genuine research orientation and a competitive moat. The more credibly a company signals that it understands AI’s dangers, the more it implicitly argues that it — rather than regulators, rivals, or open-source communities — should be trusted to manage them. Jones’s expected-utility framing, however provocative, is intellectually consistent with that posture: it treats extinction risk as a variable to be optimized, not a threshold to be avoided, which is exactly the kind of analysis that requires a trusted institutional steward to interpret responsibly. Whether Anthropic can sustain that stewardship argument while its products appear in military targeting workflows is the harder question.

How Anthropic’s Safety Stance Compares to Its Peers

Anthropic is not the only major AI developer with a stated commitment to safety, but its public positioning is distinctively intense. A comparison across the sector’s leading actors illustrates how varied the approaches — and the credibility gaps — actually are.

The table illustrates a recurring pattern across the sector: safety commitments are real but contested at the margins, and every major player has a gap between stated principles and operational decisions. What distinguishes Anthropic is the intensity of its public rhetoric — which is precisely why the Jones hire and the Claude deployment reports carry more reputational weight for this company than they might for a competitor with a lower-profile safety brand.

For executives tracking AI governance risk, this matters. Companies that build procurement or partnership strategies around a vendor’s safety credentials — a trend accelerating as enterprises move toward custom AI deployments — should treat safety claims as a starting point for due diligence, not a conclusion. The Jones episode is a reminder that the economists, researchers, and ethicists an AI company hires reflect its actual intellectual framework, not just its press releases.

There is also a regulatory dimension. As Washington’s appetite for technology regulation grows — a trend visible across both AI and crypto — companies that have cultivated a safety-first reputation with policymakers face heightened exposure when that reputation comes under scrutiny. Anthropic has been an active participant in AI policy conversations in Washington; any erosion of its credibility as a responsible actor has downstream consequences for how its lobbying and policy positions are received.

The Implications That Matter

1. Safety branding is now a competitive asset with measurable reputational risk. Anthropic’s identity is more tightly bound to AI safety than any peer; the Jones hire and the Claude deployment reports therefore carry disproportionate reputational weight for this company specifically.
2. Expected-utility frameworks applied to existential risk deserve serious scrutiny from policymakers. Jones’s paper is academically legitimate, but the leap from economic modeling to operational AI governance is not automatic — and regulators engaging with Anthropic on safety standards should understand the intellectual range inside the company.
3. Enterprise buyers should treat AI safety claims as due-diligence inputs, not guarantees. The gap between a vendor’s public safety positioning and its actual hiring, partnerships, and deployment decisions is increasingly visible and increasingly material to procurement risk.
4. The existential-risk narrative structurally benefits established players. Companies that credibly invoke civilizational stakes implicitly argue for their own indispensability — a dynamic that serves incumbents and may distort both investment flows and regulatory design, as analysis of AI bubble dynamics has noted.
5. Washington’s engagement with Anthropic as a policy partner deserves fresh scrutiny. If the company’s internal intellectual framework tolerates extinction-risk trade-offs that most policymakers would find unacceptable, the alignment between Anthropic’s policy recommendations and the public interest is worth re-examining.