A sharp new front has opened in the battle over artificial intelligence in the U.S. military, and it centers on a question that would have seemed absurd just two years ago: Is the Pentagon strong-arming AI companies into abandoning their own safety guardrails?
Senator Edward J. Markey, the Massachusetts Democrat who sits on the Commerce, Science, and Transportation Committee, sent a pointed letter to Defense Secretary Pete Hegseth on June 19, 2025, demanding answers about what he called a pattern of “intimidation” by Department of Defense officials against AI vendors who maintain ethical use policies. The letter, published on Markey’s Senate website, names no specific companies but arrives at a moment when Anthropic — the AI safety–focused startup behind the Claude family of models — finds itself at the center of an intensifying controversy over its relationship with the U.S. defense establishment.
The Senator’s Allegations: A Government Pressuring Its Own Contractors
Markey’s letter lays out a series of specific concerns. He writes that he has received reports of DoD officials pressuring AI companies to weaken or remove acceptable use policies (AUPs) that restrict how their models can be deployed in military and intelligence contexts. These policies, common across the major AI labs, typically prohibit the use of AI systems in applications involving autonomous weapons, mass surveillance, or lethal targeting without human oversight.
“I am troubled by reports that Department of Defense officials have sought to intimidate AI companies into removing or weakening their responsible use policies as a condition of doing business with the federal government,” Markey wrote. He asked Hegseth to respond to a series of questions by July 3, 2025, including whether DoD personnel have explicitly requested that vendors alter their AUPs, whether any contracts have been conditioned on such changes, and whether the department has a formal policy governing its interactions with vendors on these matters.
Anthropic: From AI Safety Standard-Bearer to Pentagon Partner
The timing of Markey’s intervention is not coincidental. As WebProNews reported in its detailed analysis, Anthropic has undergone a remarkable transformation over the past 18 months. Founded in 2021 by former OpenAI researchers Dario and Daniela Amodei, the company built its brand on the premise that it would develop AI responsibly and with an abundance of caution. Its “Constitutional AI” approach and published acceptable use policies became reference points for the industry.
But Anthropic’s posture began shifting in late 2024 and accelerated into 2025. The company quietly revised its acceptable use policy to permit certain defense and intelligence applications, a move that opened the door to Pentagon contracts. By early 2025, Anthropic had secured agreements to provide its Claude models to defense and intelligence agencies, including through intermediary platforms like Palantir and Amazon Web Services’ classified cloud infrastructure. The company argued that engagement with the national security community was preferable to ceding that ground to less safety-conscious competitors — a rationale that has become standard in Silicon Valley’s defense pivot.
The Ethical Crossroads That Won’t Go Away
What makes the Anthropic case so consequential, as WebProNews noted in a separate investigation, is that the company’s identity was forged in opposition to precisely the kind of uncritical commercialization it now appears to be embracing. Several former Anthropic employees have spoken publicly about their discomfort with the direction, and internal tensions reportedly contributed to a wave of departures from the company’s safety and policy teams in early 2025.
The broader industry is watching closely. Google, which faced its own internal revolt over Project Maven in 2018, has since re-engaged with the Pentagon under revised AI principles. OpenAI removed its blanket prohibition on military use in January 2024. Microsoft has long been a major defense contractor. But Anthropic’s case is different in kind because the company’s founding narrative was explicitly about building a safer alternative. When the safer alternative starts behaving like everyone else, the question becomes whether “AI safety” was ever more than a marketing strategy.
Markey’s Broader Concern: The Erosion of Voluntary Safeguards
Senator Markey’s letter goes beyond any single company. His concern is structural: if the federal government’s largest buyer of technology is using its purchasing power to strip away the voluntary safety commitments that AI companies have made, then the entire framework of self-regulation — already fragile — collapses. The U.S. has no comprehensive federal AI legislation. The Biden administration’s executive order on AI, issued in October 2023, established some reporting requirements and safety testing protocols, but the Trump administration has signaled a preference for deregulation and faster deployment.
In this environment, the acceptable use policies maintained by companies like Anthropic, OpenAI, Google DeepMind, and Meta represent the closest thing to binding safety standards that exists. If the Pentagon is actively working to erode those standards, Markey argues, Congress needs to know — and may need to act.
The Defense Department’s Accelerating AI Appetite
The pressure on AI vendors comes against a backdrop of rapidly expanding military AI ambitions. The DoD’s Chief Digital and Artificial Intelligence Office (CDAO) has been aggressively pushing to integrate generative AI across military operations, from logistics and maintenance to intelligence analysis and battlefield decision support. Deputy Secretary of Defense Kathleen Hicks, before her departure, had championed the Replicator initiative aimed at fielding autonomous systems at scale. The current Pentagon leadership under Secretary Hegseth has, by multiple accounts, accelerated that timeline.
Defense procurement officials have made clear in public forums that they want AI models with fewer restrictions, not more. At a defense technology conference in May 2025, a senior CDAO official reportedly told attendees that vendors whose policies “create friction” in deployment would find themselves at a competitive disadvantage. While such statements stop short of explicit coercion, they establish a clear incentive structure: companies that maintain restrictive AUPs risk losing access to what could become the largest AI procurement market in the world.
The Competitive Dynamics Driving Capitulation
For AI companies burning through billions in compute costs and facing investor pressure to demonstrate revenue, the defense market is extraordinarily attractive. Pentagon AI spending is projected to exceed $15 billion annually by 2027, according to estimates from the Center for Strategic and International Studies. For a company like Anthropic — which has raised over $10 billion but has yet to turn a profit — the financial logic of defense work is difficult to resist.
This creates what critics describe as a race to the bottom on safety. If Anthropic won’t provide unrestricted models, the Pentagon can turn to OpenAI, or to open-source alternatives like Meta’s Llama, or to defense-specific AI firms like Scale AI and Shield AI that have no comparable safety commitments. The result is a dynamic in which maintaining ethical guardrails becomes a competitive liability rather than an asset.
What Markey Wants — and What Comes Next
The senator’s letter asks seven specific questions, including whether any DoD component has “threatened, implicitly or explicitly, to exclude or disadvantage AI vendors that maintain responsible use policies,” and whether the department has conducted any legal review of the practice. He also asks whether the DoD has established any internal guidelines on engaging with vendors about their AUPs, and whether any contracts currently in force include provisions requiring vendors to modify their safety policies.
If Hegseth responds — and there is no guarantee he will, given the current administration’s posture toward congressional oversight — the answers could reshape the debate. Evidence of systematic pressure would strengthen the case for legislation mandating minimum safety standards for AI systems used in government, a proposal that has been circulating in draft form among Senate Democrats for months. Conversely, a denial from the Pentagon would put the burden back on companies like Anthropic to explain why their policies have shifted so dramatically without external coercion.
The Stakes for the AI Industry’s Credibility
The Markey letter crystallizes a tension that has been building since generative AI burst into public consciousness in late 2022. The major AI labs have consistently argued that voluntary commitments, industry best practices, and self-governance are sufficient to manage the risks of increasingly powerful AI systems. They have resisted binding regulation, arguing that it would slow innovation and cede leadership to China.
But if those same voluntary commitments can be abandoned the moment a large enough customer applies pressure, the argument for self-regulation loses its foundation. Senator Markey, who has been among the most vocal advocates for AI legislation in Congress, is effectively calling the industry’s bluff. The question now is whether anyone — in the Pentagon, in the AI labs, or in Congress — is prepared to answer honestly.
The July 3 deadline for Secretary Hegseth’s response will be closely watched. But regardless of what the Pentagon says, the underlying dynamics are unlikely to change. The U.S. military wants AI with fewer restrictions. AI companies want defense revenue. And the safety commitments that were supposed to protect the public interest are caught in between, with no legal authority to enforce them and mounting financial incentives to abandon them. That is the real story — not whether any single company made the right call, but whether the entire system of voluntary AI safety was ever built to withstand this kind of pressure.
WebProNews is an iEntry Publication