"Dario Amodei's Oppenheimer Moment" is an opinion essay by Ross Andersen, published in The Atlantic on March 11, 2026 (Source: https://www.theatlantic.com/technology/2026/03/anthropic-dod-ai-utopianism/686327/). It frames Dario Amodei through a sustained analogy to the nuclear physicists of the Manhattan Project, arguing that the creators of a dangerous dual-use technology lose meaningful control over its use once the technology works. Andersen presents the Pentagon's DPA supply-chain-risk designation against Anthropic (February 2026) as the moment that control passed from creator to state.
Summary of argument
Andersen's central claim is that Amodei is a present-day analogue of Leo Szilard and J. Robert Oppenheimer: a technologist who is simultaneously the most optimistic and the most fearful about the technology he is building, and who, in Andersen's reading, is learning that "the people who create a powerful technology don't usually get the final say in how it's used."
The essay develops this through two registers in Amodei's own writing. In the utopian register, "Machines of Loving Grace" — a 15,000-word essay — anchors a vision of a "country of geniuses" in a data center, infectious disease and cancer cured, lifespans doubled, and decades of scientific progress compressed by 2035; Andersen names Demis Hassabis and Sam Altman as fellow utopians. Andersen treats that essay as an outlier against Amodei's broader writing, which carries a more fearful register: Amodei recommends Richard Rhodes's The Making of the Atomic Bomb to staff, thinks of contemporary AI researchers as Manhattan Project analogues, and has warned that superhuman AI could be "even more dangerous than nuclear weapons."
Andersen draws explicit parallels to four figures from the nuclear era. Leo Szilard read H.G. Wells's The World Set Free (1914) the year before he envisioned the nuclear chain reaction (1933); the novel coined the term "atomic bomb" and imagined a great war delivering humanity into lasting peace. Edward Teller imagined using nuclear explosions to "redirect rivers," carve highways through mountain rock, and dig a polar-bear-shaped harbor in Alaska; the Soviets did detonate three nuclear explosives in an attempt to redirect the Pechora River to the Caspian Sea. Lewis Strauss coined "too cheap to meter" for atomic energy and promised the transmutation of elements, the end of famine, and vastly extended human lifespans; Andersen adds a parenthetical comparison that "Altman has recently riffed on this by claiming that intelligence will soon be too cheap to meter." Niels Bohr hoped the bomb's power would scare statesmen into a posture of "radical openness" and a step change in moral evolution, and Churchill said Bohr should be locked up for suggesting the allies reveal their nuclear hand.
Against those hopes, Andersen recounts what followed. The world was not set free; atomic energy did not produce flying cars or routine spaceflight, and he cites the failures of the Ford Nucleon concept car, Studebaker-Packard's force-field design, the Air Force atomic-airplane concepts, and the Dyson/Taylor Project Orion. Project Plowshare conducted 27 peaceful detonations that, in Andersen's words, "achieved nothing except contamination and the galvanization of the environmental movement." Bohr's openness proposal to FDR was rejected by Churchill, and Oppenheimer's later UN proposal was rejected by the Soviets. Andersen notes that nine nations today hold roughly 12,000 warheads and that the final treaty between the United States and Russia expired without replacement.
The essay reads the Anthropic-Pentagon episode as the contemporary instance of this pattern. Andersen describes Amodei as holding two red lines — no mass surveillance of US citizens, and no autonomous weapons that kill without human oversight — which the Pentagon refused, demanding unrestricted use and invoking a supply-chain-risk DPA designation Andersen describes as never before used against an American company. Andersen reports that Amodei's February 27 staff memo characterized OpenAI's competing deal as "safety theater" and said the OpenAI episode "revealed who they really are" (see Presidential Determination 2026-10 (Defense Production Act Section 303 — AI Grid Components) and Anthropic v. United States (Pentagon ban challenge)).
Andersen's closing point applies the nuclear lesson directly: "Their leverage was front-loaded: They could choose to create their terrible weapons or not, but once they'd successfully tested even one, they'd already forfeited it." Of Amodei, he writes that "whether anyone will be able to control such a technological force remains an open question, and either way, it certainly won't be him."
Reception and connections
The essay is the first sustained treatment anchoring the Amodei–Oppenheimer analogy that has since recurred in coverage of Anthropic. It connects Anthropic v. United States (Pentagon ban challenge) and the Presidential Determination 2026-10 (Defense Production Act Section 303 — AI Grid Components) to a broader argument about whether technologists retain control of dangerous dual-use research, and it sets the utopian register of Machines of Loving Grace against the operational reality of frontier-model deployment under state pressure. Later developments framed by Anthropic in similar "responsibility cannot be unilaterally relinquished" terms include the May 2026 Pope Leo XIV / Anthropic AI encyclical, the Mythos secrecy controversy, and the FSB briefing.
Key quotes
"Amodei seems to think of today's AI researchers as comparable to Manhattan Project scientists, and has been known to recommend The Making of the Atomic Bomb."
"Like the proposals put forth by Bohr and Oppenheimer, the treaties were defeated by the cold logic of competitive advantage, which will also likely shape the global future of AI."
"The people who create a powerful technology don't usually get the final say in how it's used."
"Their leverage was front-loaded: They could choose to create their terrible weapons or not, but once they'd successfully tested even one, they'd already forfeited it."
Provenance
Author: Ross Andersen (The Atlantic). Published March 11, 2026. Source URL: https://www.theatlantic.com/technology/2026/03/anthropic-dod-ai-utopianism/686327/. As an opinion essay, its claims are presented as Andersen's argument rather than as independently established fact.
Relationships
- supports: Anthropic v. United States (Pentagon ban challenge) — provides the historical-analogy frame
- supports: AI and National Security — argues democratic control of frontier AI may not be sustainable under competitive-pressure dynamics
- contradicts: Machines of Loving Grace — frames the utopian register as historically naive against the structural lesson of nuclear
- related: Dario Amodei, Presidential Determination 2026-10 (Defense Production Act Section 303 — AI Grid Components), Autonomous Weapons, AI and Surveillance, Dual-Use Frontier AI, AI Race Dynamics