The hearing room was a converted office in a temporary building near the Washington Monument, the kind of government annex that smelled of floor wax and radiator dust. On April 12, 1954, J. Robert Oppenheimer walked into it and sat down across from a three-member panel that would decide whether the man who had built the atomic bomb was too dangerous to be trusted.

The charges were not espionage. They were not treason. The Atomic Energy Commission’s personnel security board had been asked to determine whether Oppenheimer constituted a “security risk,” a bureaucratic designation that had, until that moment, been reserved for suspected foreign agents and ideological subversives. The physicist who had directed the Manhattan Project, who had handed the United States the most decisive military advantage in human history, was now being processed through the same administrative machinery designed to catch Soviet spies.

The real issue was the hydrogen bomb. In 1949, when the Soviet Union detonated its first atomic weapon and the question arose whether America should pursue a thermonuclear “Super” bomb a thousand times more powerful, Oppenheimer’s General Advisory Committee had voted unanimously against a crash program. Oppenheimer and Harvard president James Conant went further, writing in an addendum that “we believe a super bomb should never be produced,” calling it a weapon of genocide that offered no meaningful strategic advantage over the arsenal the country already possessed.

President Truman overruled them. The hydrogen bomb was built. And Lewis Strauss, the AEC chairman who had championed the Super, never forgave Oppenheimer for opposing it.

The security hearing that followed was, by the admission of participants on both sides, less a fact-finding proceeding than a prosecution. The government’s attorney, Roger Robb, had access to Oppenheimer’s wiretapped conversations. Oppenheimer’s own lawyers were denied security clearances to review classified evidence against their client. Over four weeks of testimony, forty witnesses were called. The fundamental question was never whether Oppenheimer had betrayed his country. Everyone agreed he had not. The question was whether his policy disagreements and his personal associations made him a “risk.”

The panel voted 2-1 against him. Gordon Gray and Thomas Morgan found that Oppenheimer’s “conduct and associations reflected a serious disregard for the requirements of the security system.” Ward Evans, the lone dissenter, wrote that denying clearance to someone who had been cleared in 1947 “when we must know he is less of a security risk now than he was then, seems to be hardly the procedure to be adopted in a free country.” He called the decision a “black mark on the escutcheon of our country.”

The full AEC commission voted 4-1 to sustain the ruling. Henry DeWolf Smyth cast the only dissenting vote. Oppenheimer, found unanimously loyal by every panel that reviewed his case, was stripped of his clearance and excluded from the corridors of power he had helped build.

The scientific community was stunned. Albert Einstein, who had told Oppenheimer he was a fool for even answering the charges, thought he should have resigned rather than submit. Physicists across the country understood the unspoken message: the government’s most powerful creator had been punished not for disloyalty, but for the temerity of telling the government that some applications of his creation were too dangerous to pursue.

Oppenheimer spent his remaining years as director of the Institute for Advanced Study in Princeton, his influence on nuclear policy severed. He died in 1967. In December 2022, sixty-eight years after the hearing, the Department of Energy formally vacated the AEC’s decision, declaring that it had been the result of a “flawed process” that violated the commission’s own regulations.

The Echo

On March 5, 2026, the Department of Defense officially designated the artificial intelligence company Anthropic a “supply chain risk,” a classification that had never before been applied to an American company. The label, historically reserved for entities like Huawei and Kaspersky Lab, requires defense contractors to certify that they do not use Anthropic’s Claude AI models in any work with the military.

The dispute began in July 2025, when Anthropic signed a $200 million contract to deploy Claude on the Pentagon’s GenAI.mil platform. By September, negotiations stalled. The DOD wanted unrestricted access across all lawful purposes. Anthropic wanted contractual assurance that its technology would not be used for fully autonomous weapons or domestic mass surveillance. The two could not agree. President Trump then ordered all federal agencies to cease using Anthropic’s technology, and Defense Secretary Pete Hegseth applied the supply chain risk designation.

Anthropic sued in two courts. A federal judge in San Francisco, citing “First Amendment retaliation,” granted an injunction blocking the ban on non-defense government use. A federal appeals court in Washington denied Anthropic’s request to block the Pentagon blacklisting, ruling that “judicial management of how the Department of War secures vital AI technology during an active military conflict” outweighed the company’s financial harm. The split decisions left Anthropic excluded from defense contracts but permitted to serve other agencies.

The tech industry rallied behind Anthropic, much as the scientific community had rallied behind Oppenheimer. The structural parallels are difficult to ignore. In both cases, the creator of the most consequential technology of the era built the thing, then tried to impose limits on how it could be used, and was punished through an administrative designation normally reserved for foreign adversaries.

Supporters of the government’s position make a case that deserves serious engagement. The United States is in an active military conflict. Its principal AI competitor, China, imposes no such restrictions on military applications. A company that signs a $200 million defense contract and then dictates terms on deployment may genuinely compromise operational capability. The DOD’s lawyer argued in court that officials had “come to worry that Anthropic may in the future take action to sabotage or subvert IT systems,” a concern rooted in the company’s insistence on maintaining the ability to restrict access.

Anthropic’s defenders counter that the supply chain risk label is a political weapon, not a genuine security assessment. The designation was applied after a policy disagreement, not after any evidence of sabotage or foreign compromise. The California judge who reviewed the evidence found “strong evidence of First Amendment retaliation.” And the precedent is alarming: if refusing to grant a government agency unconditional access to a technology makes a company a national security threat, the designation has been emptied of meaning and refilled with something far more dangerous.

Both arguments have weight. And the Oppenheimer case offers no clean resolution. In 1954, the government’s position seemed unassailable: the Soviets had the bomb, the Cold War was escalating, and Oppenheimer was obstructing the national defense. Sixty-eight years later, the Department of Energy vacated the decision, calling it a “flawed process.” But the hydrogen bomb that Oppenheimer tried to prevent became the cornerstone of deterrence that kept the Cold War from becoming a hot one. He may have been right about the ethics and wrong about the strategy. Or the reverse. History has not settled the question.

What history does offer is a pattern. When a powerful new technology emerges, the tension between its creator and the state over who controls its application is not a bug. It is a feature of every era that has produced a weapon too dangerous to use freely and too important to leave ungoverned. The atomic bomb. Encryption. Artificial intelligence. Each time, the same argument plays out: security demands access, conscience demands limits, and the tools of bureaucracy get pressed into service on both sides.

The Oppenheimer hearing ended with a vote. The Anthropic dispute is still in court. In both cases, the deeper question remains unanswered.

Who gets to decide what the most powerful technology of an era is allowed to do?