Trump pressures AI safety rules, firms scramble for Pentagon

Anthropic’s safety-focused brand was tested in March 2026 when the Trump administration declared it a supply chain risk, ordered the federal government to stop using Claude, and pushed the Pentagon toward OpenAI within hours—while Anthropic later scrapped bind

In March 2026, Anthropic’s safety reputation was supposed to be its differentiator. The company had been built in 2021 with detailed safety measures, and it marketed those safeguards as a corporate standard—especially after researchers left a rival they had criticized for moving too fast.

Then the Trump administration stepped in.

The administration declared Anthropic a supply chain risk and ordered the federal government to stop using Anthropic and its large language model, Claude, branding the company a national security risk. Within hours, OpenAI struck a deal to become the Pentagon’s supplier.

The pressure wasn’t abstract. Anthropic had refused to remove safeguards it built into products it had supplied to the Pentagon—safeguards that prohibited domestic surveillance and autonomous weapons. The administration’s move framed those boundaries as a threat. Yet at the same time, just after the Pentagon banned Claude, U.S. military use continued: the technology was still being used to select and target sites to bomb in Iran.

Inside Anthropic, the conflict with Trump also left a trail the public narrative didn’t show. During its clash with the administration, the company quietly scrapped the binding principles in its main safety policy. Several weeks earlier. Anthropic’s head of safeguards research had resigned. warning that “the world is in peril.” And a week after the Pentagon officially banned Claude. the military was still deploying the technology in operations tied to bombing in Iran.

There’s a second irony in the timeline: the policy fight was treated as a question of compliance and ideology, while the practical question—who gets to keep guardrails when government contracts are on the line—became a matter of immediate business survival.

That survival instinct can reshape decisions even for companies that still take safety work seriously. In the midst of this scramble. OpenAI’s CEO Sam Altman told his board of directors the deal looked “opportunistic and sloppy.” The company moved forward anyway. because admitting an action looks bad is not the same as being willing to fall behind.

A single U.S. Defense Department AI contract can be worth billions of dollars. It can also provide access to data no private company could otherwise have, and it can unlock further government work. In that environment, companies face a stark tradeoff: maintaining ethics guardrails can mean ceding ground to competitors that remove them.

This is where the facts point toward a repeating outcome. The situation can be read like the classic prisoner’s dilemma: if one company maintains safety provisions while another strips them away. the company that discards safeguards gains contracts and advantage. If both maintain the provisions, the guardrails might survive. But neither company can be sure the other will hold the line—and being left behind is not a good option. The rational choice, under uncertainty, becomes discarding safety measures.

The difference from an ordinary race to the bottom is that the trap isn’t accidental. The incentive structure is being maintained by government action.

Palantir is cited as a parallel case in how companies can “solve” that dilemma by defecting first. The data analytics firm was founded by Peter Thiel and run by Alex Karp. who spent years denouncing “woke” Silicon Valley. Palantir built a business model around government surveillance and military data infrastructure. While Palantir has said it is committed to privacy and civil liberties, critics contend the company is dismantling those protections.

Under the Trump administration, Palantir’s stock has surged, its contracts have expanded, and the company has a front-row seat where AI policy is being written.

The argument behind these examples lands on a single question that governs what happens next: when safety teams are weakened across the industry—such as OpenAI’s Superalignment team and Microsoft’s ethics unit—it isn’t presented as a decision to abandon safety. It’s described as a pattern of collective. incremental compromises that quietly reorient what “safety” means. away from the public and toward the state. The harm and risks, the author notes, fall on everyone whose lives are shaped by AI systems.

The administration’s framing also matters. The Trump approach has argued that AI safety standards and user restrictions are ideological impositions rather than sound engineering decisions. The “Preventing Woke AI” executive order of July 23. 2025. is described as not changing what companies are allowed to do with their products—but the label attached to basic ethics protections is said to make those protections politically costly to maintain.

Legal and contracting pressure is part of the mechanism. The Brennan Center, a legal policy and advocacy organization, has documented how AI ethics is being redefined through contract negotiations. In those cases. the government weaponizes terms such as biased to disqualify companies that maintain civil rights protections from competing for federal contracts.

From there, the definition of safety shifts again—this time in government documents and contract language. The original definition of AI-related safety is described as moving from protecting the public toward making systems controllable for the state. The “anti-woke” framing accelerates this shift by making ethics requirements look ideological rather than technical. Removing them can then be described not as a safety reduction but as a correction.

The piece argues that this does not require companies to act in bad faith. Safety teams can still be doing rigorous work. Companies may not be lying when they describe their safety commitments; those commitments are simply oriented toward the government rather than the public.

The political stakes are laid out in a hard contradiction: the case for stronger AI regulation assumes a government constrains commercial entities on behalf of the public. But blacklisting a company for maintaining civil rights protections. and then banning the military deployment of its AI hours later. is described as showing that the federal government in this instance enables the harm that regulation is meant to prevent.

Expanding regulatory authority over AI companies, the author concludes, does not necessarily protect citizens. In authoritarian regimes, safety regulations intended to constrain corporate power can become tools to coerce compliance.

Michael Gregory is an assistant professor of philosophy at Clemson University.

Anthropic Claude OpenAI Pentagon AI safety national security risk supply chain risk prisoner’s dilemma defense contracts Sam Altman Boeing? no Iran targeting Superalignment team ethics unit Brennan Center Preventing Woke AI July 23 2025 Palantir Alex Karp Peter Thiel civil liberties rule of law