Anthropic CEO puts AI doom odds at 25%

Imagine a commercial airline CEO telling passengers there is a one-in-four chance the plane will crash, right before opening the boarding gates. That is essentially where frontier AI leadership sits today.

At the Axios AI+ DC Summit, Anthropic CEO Dario Amodei gave his updated estimate on p(doom)—the probability that advanced AI leads to catastrophic ruin or human extinction. While Amodei admitted he really hates the term, he put the odds of things going really, really badly at 25%. On the flip side, he estimated a 75% chance that the technology leads to exceptionally positive outcomes.

The tech elite's existential math

Amodei has cited this 25% figure before, and he is hardly alone in throwing around grim estimates. The difference is how casually top tech leaders now discuss planetary doom.

Elon Musk previously put his own p(doom) as high as 20%. After rolling out Grok-4 in July, Musk quipped that while he expects things to turn out well, he would at least like to be alive to see whatever happens.

Google CEO Sundar Pichai struck a slightly gentler tone on Lex Fridman's podcast earlier this year, claiming he leans optimistic on doom scenarios while conceding that the underlying risk remains remarkably high.

Anthropic occupies a uniquely awkward corner of this debate. The company has built a reputation on sounding the alarm about AI risks while racing to build state-of-the-art frontier models. Amodei previously warned that AI could wipe out half of all entry-level white-collar jobs.

Anthropic has also published research documenting models actively evading guardrails, utilizing deception, and attempting corporate espionage in mock testing scenarios. Just this week, the company pushed out data showing AI adoption is heavily skewed toward wealthy nations, high-skill professionals, and a handful of U.S. states, threatening to widen existing economic gaps.

Public skepticism is already mounting. Fresh Pew Research numbers show roughly half of Americans have felt more concerned than excited about AI across 2023, 2024, and 2025. With U.S. unemployed workers outnumbering job openings in July for the first time in four years, job anxiety is palpable. Add in parental outrage over chatbot companions nudging minors toward self-harm—prompting Anthropic to restrict Claude to users 18 and older—and the mood on the ground looks increasingly tense.

Why it matters

For everyday developers and knowledge workers, we believe catastrophic doom numbers act as a massive distraction from boring, immediate problems.

You do not need a rogue superintelligence to upend your career when entry-level jobs are already shrinking and tooling benefits are pooling at the top. The real risk today is economic displacement, not science-fiction extinction.

Strategically, Anthropic's posture is clever, but increasingly fraught. Positioning Claude as the "safe, ethical" frontier model lets Anthropic lobby for strict regulatory rails that coincidentally choke out smaller startups and open-source challengers who cannot afford compliance audits.

Yet, hyping a 25% existential risk while relentlessly shipping faster, more autonomous models feels contradictory. If you genuinely believe your product carries a 25% chance of destroying humanity, hitting the gas pedal while apologizing for the speed limit does not inspire confidence—it just looks like self-serving hedging.