A Manifesto That Changed History
In 1955, at the height of the Cold War, Albert Einstein and Bertrand Russell did something that scientists rarely do: they issued a political ultimatum to the entire human race. Their manifesto did not merely warn of nuclear fallout. It posed a question so stark it has never been bettered: “Shall we put an end to the human race, or shall mankind renounce war?”
It was, in every meaningful sense, a scientific mutiny — and it gave birth to the modern arms control movement.1
The signatories were not firebrands. They were the most eminent scientists of the age: Max Born, Percy Bridgman, Leopold Infeld, Frederic Joliot-Curie, Herman Muller, Linus Pauling, Cecil Powell, Joseph Rotblat, and Hideki Yukawa. They acted not because they were naive about power, but because they understood it with a precision that no politician of the era could match. They had built the weapon. They knew what it could do. And they refused to be silent.
That manifesto produced the Pugwash Conferences on nuclear disarmament, shaped the Partial Test Ban Treaty of 1963, and established the principle — now embedded in arms control law — that scientific expertise confers not merely capability but responsibility.
We are overdue for its twenty-first-century equivalent.
Hawking’s Warning and the Silence That Followed
When Stephen Hawking issued his landmark warning in December 2014 — that the development of full artificial intelligence “could spell the end of the human race” — it was received with the same paradoxical mixture of global fanfare and immediate strategic neglect that had greeted Einstein and Russell nearly sixty years earlier.
Hawking was not speaking loosely. He was pointing to a structural danger: that an intelligence capable of redesigning itself at a rate no human institution could track would, by definition, become ungovernable. He returned to the theme repeatedly in the years before his death in 2018, warning in a 2017 interview that AI “could be the worst event in the history of our civilization” if not properly managed, and calling for new international regulatory bodies analogous to those built around nuclear and biological weapons.2
The wizards of Armageddon — to borrow Fred Kaplan’s phrase — heard Hawking perfectly well. They simply waited for the technology to mature before moving to capture it.3
That moment has now arrived.
The Standoff
On March 4, 2026, the Trump administration designated Anthropic a “Supply-Chain Risk to National Security.” The label has historically been reserved for hostile foreign actors — Chinese telecom equipment manufacturer Huawei being the paradigm case. Its application to an American company, founded by Americans, for the explicit purpose of building AI that is safe by design, represents something genuinely new: the weaponization of a national security designation as an instrument of ideological compulsion.4
Anthropic’s position is not esoteric. The company’s founder and CEO, Dario Amodei, has drawn two clear lines: Claude, their flagship AI system, will not be deployed for mass domestic surveillance, and it will not be integrated into fully autonomous lethal weapons systems — including, critically, weapons use decision-making. These are not anti-military positions. They are the minimum conditions under which any responsible actor should be willing to release a powerful general-purpose AI into a weapons context.
The administration’s response has been characteristically blunt and comprehensively reported by the New York Times. Defense Secretary Pete Hegseth made social media postings over Anthropic’s stance, calling out “arrogance and betrayal,” and declared that “America’s warfighters will never be held hostage to the ideological whims of Big Tech.”5 Pentagon CTO Emil Michael invoked the language of pathology, accusing Amodei of harboring “a God-complex.” The President himself reportedly threatened to use “the full power of the presidency” to compel compliance. The sequence was deliberate: on February 27, the President announced on Truth Social that he was moving to ban Anthropic from all government funding. The Pentagon formally notified the company on March 4 and made the designation public on March 5." The entire interaction between former Uber executive and current Defense Department Chief Technology Officer Emil Michael is reported in the New York Times article on the Defense Department's failure to negotiate with Anthropic.6
This is not a policy disagreement. It is an attempt to conscript a private company — and the scientists who work for it — into a military application they have determined to be unsafe. It is, in the most precise sense, the suppression of scientific conscience.
The Vietnam Parallel
The precedent is uncomfortable but exact. During the Vietnam War, the Pentagon’s insistence on “operational flexibility” — its bureaucratic euphemism for the use of Agent Orange and napalm — eventually generated a rupture between the American military establishment and the academic institutions that had previously supplied it with research, talent, and legitimacy. The slow-burning scandal of toxic defoliants, of villages incinerated by jellied petroleum, became impossible to quarantine from the labs and lecture halls. One by one, universities severed ties with Pentagon-funded research programs. The scientific community, which had been complicit in the industrialization of warfare since 1945, drew a new line.
What is being demanded of Anthropic now is structurally identical to what was demanded of those institutions then: strip away the ethical constraints, hand us the capability, and trust us to use it wisely. The history of Agent Orange is the answer to that request.
The difference, of course, is scale. Agent Orange contaminated a country. An AI system, integrated without safeguards into nuclear command-and-control decision-making, could, in the wrong circumstances, contaminate a civilization.
The Two-Front Assault
The Anthropic confrontation does not stand alone. It is one face of a two-front assault on the architecture of restraint that the post-war world laboriously constructed.
The other front is nuclear testing. The administration is simultaneously pressing for the resumption of explosive nuclear tests, broadly accepted as technically unneeded and unnecessary — they would be the first since the United States signed on to the Comprehensive Nuclear-Test-Ban Treaty framework in 1996. The CTBT has never entered into force, but the moratorium it represents has (with the exception of North Korea) held for three decades. Breaking it now, at precisely the moment when AI is being proposed as a component of weapon-use decision-making, is no coincidence. It is the completion of a logic: if you automate the trigger and remove the human from the loop, you also need to be confident the warhead performs as designed. Testing and AI autonomy are two faces of the same strategic posture.
As Amodei himself has noted: “Using these systems for mass domestic surveillance is incompatible with democratic values... we will not knowingly provide a product that puts America’s warfighters and civilians at risk.”
The danger Amodei is gesturing toward — though the administration prefers not to name it — is the possibility of AI-driven escalation cycles so rapid that no human decision-maker could intervene before a conventional exchange became nuclear. Military strategists have a term for the nightmare scenario: a Flash War. It has been a theoretical concern since the early days of automated early-warning systems. Adding large language model-driven decision support to nuclear command structures — without safeguards, without human-in-the-loop requirements — transforms a theoretical concern into a design specification.
The Industry’s Diluted Response
Here, the news is mixed, and the mixture is instructive.
A coalition of researchers and engineers — including [verify: approximately 37 individuals from OpenAI and Google, reportedly including Google Chief Scientist Jeff Dean — filed an amicus brief supporting Anthropic’s position, warning of a “chilling effect” that would silence precisely the experts most qualified to identify and resist dangerous applications. That is a meaningful act of professional solidarity, and it deserves recognition.
The executive leadership of those same companies is a different matter. Google, Meta, and OpenAI have maintained a conspicuous public silence — eyeing, one assumes, the federal contracts that now hang in the balance. If the safety departments of these firms exist for any purpose, it is for moments like this one. Their silence transforms those departments from a genuine function into a reputational accessory.
The Einstein-Russell Manifesto worked because the most eminent scientists of the age signed it, not their graduate students. The industry’s researcher-level response to the Anthropic confrontation is admirable. Its executive-level response is a moral abdication, and it should be named as such.
A Campaign of Conscience
The Einstein-Russell Manifesto did not produce results because it was eloquent. It produced results because it was followed by organized, sustained, internationally coordinated pressure — the Pugwash movement, the scientific boycotts, the Congressional testimony, the years of painstaking treaty negotiation. Moral clarity requires institutional expression to become policy.
The same logic applies here. Three actions are available to anyone reading this:
Pressure the industry’s executive leadership. Google, Meta, and OpenAI should be asked directly and publicly whether they endorse the principle that AI companies have the right to maintain safety guardrails — or whether their silence in this case signals otherwise. The question should be posed to their boards, their investors, and their public affairs offices. Quiet compliance, in this context, is not neutrality. It is endorsement.
Engage your Congressional representatives. The weaponization of “Supply-Chain Risk” designations as an instrument of viewpoint-based retaliation against American companies is a legitimate legislative concern, entirely separable from one’s views on AI regulation. Ask your senators and representatives to investigate whether the Anthropic designation meets the legal standard the label was designed to enforce — and to oppose any funding for the resumption of explosive nuclear testing. The message is simple: a Human in the Loop is a security requirement, not an ideological preference.
Support the watchdog institutions. The Union of Concerned Scientists, the Federation of American Scientists, The Bulletin of the Atomic Scientists, the Nuclear Threat Initiative, and the Arms Control Association have been among the first to identify what is actually at stake in the Anthropic confrontation. They are the institutional heirs to the Einstein-Russell legacy, and they are doing the work in Washington that this moment requires.
The Einstein-Russell Manifesto was issued seventy years ago. Its central question: shall we put an end to the human race, or shall mankind renounce war? — has not been answered. We have only deferred it through the patient construction of treaties, moratoriums, and safeguards that are now being dismantled, one designation at a time.
Dario Amodei did not ask to become a test case. But the stand he has taken — that there are applications of AI to which no responsible developer should consent — is exactly the kind of scientific conscience that the Einstein-Russell moment called for. The question is whether his industry, and his country, will meet him there.
© 2026 Farooq Hussain
Notes
The Einstein Russell Manifesto, Pugwash Conferences on Science and World Affairs
Stephen Hawking warns artificial intelligence could end mankind, BBC, December 2, 2014, Stephen Hawking, Keynote Address, 2017 Web Summit, YouTube
Kaplan, Fred, M., Wizards of Armageddon, Simon and Schuster, 1983
See Truth Social Post @realDonaldTrump
See Common Cause Letter Pete Hegseth vs Anthropic
How Talks Between Anthropic and the Defense Dept. Fell Apart, New York Times, March 1, 2026