Two decades before he became the chief executive of Anthropic, Dario Amodei warned that scientists must not become “amoral technicians” operating the machinery of war. Today, his company sits at the centre of a fierce debate over artificial intelligence, military power and the limits of corporate responsibility.
In March 2003, as the United States prepared to invade Iraq, a physics student at the California Institute of Technology was asking questions that would become increasingly relevant as artificial intelligence transformed modern warfare.
The student's name was Dario Amodei. Today, he leads Anthropic, one of the world's most influential artificial intelligence companies and the developer of the Claude family of AI models.
At the time, Amodei was not a technology executive negotiating contracts with military agencies. He was a university student who believed his fellow scientists had a responsibility to engage with political decisions that could shape the future of humanity.
On March 3, 2003, he published an opinion column in Caltech's student newspaper, The California Tech, titled “Anti-War Protest: Most Students Apathetic Towards World Politics.” The article criticised his fellow students for remaining disengaged as the Bush administration moved towards war with Iraq. <Cite refs={["turn287501search4","turn287501news0"]} />
His argument was not simply that the invasion was wrong. It was that educated people could not excuse themselves from political responsibility by retreating into their technical work.
More than two decades later, those words carry a different weight. Amodei now leads a company whose technology is used in national-security work, including military applications. Anthropic has publicly opposed certain uses of its AI, particularly fully autonomous weapons and mass domestic surveillance, while maintaining that AI can play a legitimate role in national defence.
The resulting tension raises an uncomfortable question: what happens when someone who once warned against the machinery of war becomes the leader of a company building technology that military institutions increasingly want to use?
The student who challenged America's march to war
Amodei's March 2003 column began with a warning from social critic Neil Postman about the danger of a society becoming so distracted by trivialities that political debate and public responsibility lose their importance.
At Caltech, Amodei argued, students were absorbed in problem sets, computer games and campus concerns while a major international crisis unfolded.
He was particularly frustrated by the gap between the number of students who expressed opposition to the invasion and the much smaller number who participated in organised protests.
The Peaceful Justice Coalition, a campus anti-war organisation, reportedly had several hundred people on its interest list, but only a handful were consistently active.
For Amodei, this was a failure of citizenship. Opposition that never translated into action offered little resistance to decisions with enormous consequences.
His criticism remains relevant far beyond the university setting. People can oppose a war, surveillance or abuses of state power in principle while continuing with their daily lives as though those decisions have nothing to do with them.
Amodei argued that scientists had a particular responsibility because their knowledge could influence the direction of public policy, technology and military power.
He warned against allowing scientists to become detached technicians who treated the machinery of war with the same casual indifference as routine calculations. <Cite refs={["turn287501search4","turn287501news0"]} />
That argument is especially striking today, when software engineers, AI researchers and technology executives can help build systems that influence intelligence analysis, military planning and the speed at which governments make consequential decisions.
His warning about Iraq proved prescient in an important respect
Amodei's column challenged the case for invading Iraq before the war began on March 20, 2003.
He questioned claims linking Saddam Hussein's government to al-Qaeda and argued that an invasion could produce consequences that outweighed its stated objectives. He also cited warnings that a war could strengthen extremist recruitment rather than eliminate the threats invoked to justify it.
The subsequent history of Iraq demonstrated how difficult it is to predict and control the consequences of military intervention.
The invasion toppled Saddam Hussein but was followed by years of insurgency, sectarian violence and instability. The emergence of the Islamic State was shaped by multiple factors, including the destruction of Iraqi state institutions, sectarian politics, the Syrian conflict and the organisation's own strategic decisions. It would be too simplistic to attribute that outcome to one decision or one individual.
Nevertheless, Amodei's underlying warning—that military intervention can create new security problems even when its proponents promise to resolve existing ones—remains an important lesson.
His student column was not a comprehensive prediction of everything that followed. It was an argument against treating war as a straightforward technical solution to a political problem.
That distinction matters because contemporary AI systems are increasingly being developed and deployed in precisely the kinds of environments where technical capability can be mistaken for strategic wisdom.
The transformation from anti-war activism to AI leadership
Amodei's public career subsequently took a different direction. He pursued scientific research, worked in the biotechnology sector and became involved in the development of advanced AI systems before co-founding Anthropic with other former OpenAI employees in 2021.
Anthropic built its reputation partly around AI safety and the principle that increasingly capable systems require safeguards against misuse.
But AI safety is not identical to pacifism. A company can believe that AI should not independently select and attack targets while also supporting its use for intelligence analysis, cybersecurity, military logistics or defensive operations.
This distinction lies at the heart of Anthropic's current position.
In a public statement on February 26, 2026, Amodei said the company had worked proactively to deploy its models within the US Department of War and intelligence community. He identified applications including intelligence analysis, modelling and simulation, operational planning and cyber operations.
At the same time, he said Anthropic would not knowingly provide a product for fully autonomous weapons under its existing restrictions, arguing that frontier AI systems were not yet reliable enough to make life-and-death targeting decisions without human involvement. He also maintained a restriction on mass domestic surveillance. <Cite refs={["turn889985search2","turn889985search1"]} />
This is not a position of rejecting every military application. It is an attempt to distinguish between uses the company considers acceptable and uses it considers too dangerous.
That makes the modern dilemma more complicated than a simple story of an anti-war activist abandoning his principles.
The central question is whether the boundaries he now defends are sufficient—and whether a private technology company can reliably enforce them once its products become embedded in national-security systems.
The Pentagon dispute exposes the limits of corporate control
In early 2026, Anthropic's relationship with the Pentagon deteriorated over the company's restrictions on certain military uses of its technology.
The dispute centred on whether the government should be able to use Claude for any lawful purpose, including applications that Anthropic's safeguards restricted.
Amodei argued that fully autonomous weapons and mass domestic surveillance presented distinct risks. The Pentagon, meanwhile, sought broader authority to use AI within its military operations.
On March 4, 2026, Anthropic said it had been designated a supply-chain risk to US national security. Amodei announced that the company would challenge the designation in court. The dispute illustrated how difficult it can be for a technology supplier to maintain restrictions when its government customer considers those restrictions incompatible with its operational requirements. <Cite refs={["turn889985search1","turn889985news9"]} />
The conflict also exposed a deeper problem for the AI industry.
A company may develop safeguards, write responsible-use policies and establish contractual limits. But once its technology is deployed within a complex military environment, the company may have limited visibility into every use of the system, particularly when information is classified or integrated into wider operational workflows.
That does not mean safeguards are pointless. It means their effectiveness depends on technical controls, contractual enforcement, monitoring and the willingness to refuse business when essential conditions cannot be maintained.
The dispute also demonstrates that the military's relationship with AI companies is not merely about buying software. It is about who ultimately controls the uses of increasingly powerful technology.
Does military AI contradict Amodei's anti-war principles?
The answer depends on what those principles require.
If opposition to war means refusing every form of military collaboration, Anthropic's work with the Pentagon would represent a fundamental contradiction.
But Amodei's 2003 column focused on the dangers of an unnecessary invasion, political disengagement and the moral responsibility of scientists. It did not establish that he believed all military technology was inherently illegitimate.
His current position is that AI can support national defence while remaining subject to restrictions on particularly dangerous applications.
Critics can reasonably ask whether intelligence analysis and operational planning help military organisations conduct wars more effectively, even when the software does not autonomously select targets. Better analysis can support defensive operations, but it can also make offensive operations more efficient.
That is the dual-use problem: the same technology may support activities with very different consequences, depending on the mission, the operator and the surrounding political decisions.
It is therefore insufficient to evaluate an AI company's ethics solely by whether it prohibits fully autonomous weapons. The broader questions include which military programmes it supports, how its tools influence decisions, what safeguards are enforceable and whether the company can withdraw support when its conditions are violated.
Equally, the mere fact that a company's technology is used by a military does not prove that its leadership has abandoned its principles. The relevant evidence lies in the specific applications, restrictions and decisions the company makes.
The danger of becoming an amoral technician
The most enduring idea in Amodei's student column was the warning that scientists should not become detached from the consequences of their work.
That concern has become more urgent as AI systems move from producing text and answering questions to supporting decisions in high-stakes environments.
An engineer developing an algorithm may see a technical challenge. A military organisation may see an opportunity to accelerate intelligence processing or planning. A political leader may see a way to achieve a strategic objective. The civilians affected by the resulting operation experience the consequences rather than the technical elegance of the system.
Each participant has a different view of the same technology.
The ethical challenge is to prevent those differences from becoming an excuse for avoiding responsibility. A company cannot settle every question about war, but it must take seriously the foreseeable uses of its products and the influence its technical decisions can exert.
Human responsibility cannot simply disappear into a chain of software developers, executives, military customers and automated systems.
If an AI system makes military operations faster, its creators should ask not only whether it works as designed but also what kinds of decisions it enables, which safeguards remain effective and who is accountable when things go wrong.
Those questions are not arguments against all military technology. They are requirements for taking its consequences seriously.
The lesson extends beyond Anthropic
Amodei's story reflects a broader struggle within the technology industry.
AI companies increasingly want to be regarded as responsible developers while also competing for lucrative government contracts and recognition as strategic national assets. Governments want the advantages of advanced AI but may resist restrictions imposed by private suppliers.
This creates an unavoidable tension between commercial growth, national-security priorities and ethical commitments.
The same challenge extends beyond the United States. Governments around the world are considering how to use AI in policing, intelligence, border control and military operations. Countries with fewer resources may face additional risks if they adopt systems developed abroad without sufficient transparency, local oversight or control over their deployment.
For African countries, the lesson is not simply to accept or reject foreign AI products. It is to develop the institutional capacity to evaluate them, establish clear procurement conditions, protect sensitive data and ensure that decisions affecting citizens remain accountable.
Technological sovereignty requires more than access to advanced tools. It requires the ability to determine how those tools are used and to challenge their deployment when public interests are at stake.
Twenty-three years later, the question remains
The contrast between Amodei's student activism and his present role is compelling because the underlying issue has not disappeared.
In 2003, he argued that scientists should not detach their work from the political consequences of war. Today, he leads a company whose products can contribute to national-security operations, while attempting to draw limits around how those products are used.
His refusal to accept certain Pentagon demands shows that the company has been willing to risk a major government relationship over its restrictions. Its continued military work, however, demonstrates that those restrictions are boundaries within a military partnership, not a rejection of military AI altogether.
Both facts matter.
The strongest test of Amodei's principles will not be whether his views in 2003 resemble his statements today. It will be whether Anthropic's present decisions consistently reflect the responsibility he described as a student—and whether its safeguards meaningfully constrain the uses he considers unacceptable.
Technology does not become ethical simply because its developers use the language of safety. Nor does every military application automatically prove that a company has abandoned its values.
The real test is what happens when ethical commitments collide with powerful customers, commercial incentives and the demands of war.
In 2003, Amodei warned against scientists becoming amoral technicians who operate the machinery of war without considering its consequences.
More than two decades later, that warning has become a question he must answer not only as a scientist, but as the leader of one of the world's most influential AI companies.
Comments
Post a Comment