Trump Administration Drafts Strict AI Contract Rules as Pentagon Clashes with Anthropic

The draft rules, unveiled in a briefing to senior defense officials last week, would require any contractor that supplies AI‑enabled systems to the Pentagon to obtain a “technology‑risk clearance” before a contract can be awarded, according to PYMNTS.com. The clearance process would examine everything from data provenance to model interpretability, and would give the Department of Defense a formal veto power over projects that it deems “unacceptable risk.” The move comes after a high‑profile clash with Anthropic, the San Francisco‑based AI startup whose Claude Opus 4.6 model was recently touted by ZDNet as capable of handling “complex, end‑to‑end enterprise workflows” with minimal human oversight. Anthropic’s push to sell the model for defense‑related use sparked a dispute over whether the company had adequately disclosed the model’s training data sources and potential bias, prompting the administration to tighten its oversight framework.

Pentagon officials have warned that the new regime could slow the procurement pipeline, but they argue that the trade‑off is necessary to avoid “unintended consequences” in mission‑critical systems. A senior defense acquisition officer, who spoke on condition of anonymity, told PYMNTS.com that the agency has already seen “several instances where contractors pushed AI capabilities into weapons platforms without a full risk assessment.” The draft also calls for periodic audits of deployed models and mandates that any updates to an AI system receive fresh clearance, a provision that mirrors recent congressional calls for greater transparency in federal AI spending.

Anthropic, meanwhile, has defended its position by highlighting the technical merits of Claude Opus 4.6. In a ZDNet feature, the company’s engineers described the model as “frontier‑level” and capable of “nailing work deliverables on the first try,” emphasizing its ability to automate tasks that traditionally required human supervision. The firm has not publicly responded to the administration’s draft rules, but its recent product announcements suggest it is still courting large‑scale enterprise customers, including those in the defense sector. The tension underscores a broader industry debate: whether rapid AI innovation can coexist with the stringent safety standards demanded by national security stakeholders.

Industry observers note that the Trump administration’s approach marks a departure from the more permissive AI procurement policies of the previous administration, which favored “fast‑track” contracts to keep pace with China’s AI investments. According to PYMNTS.com, the new draft is the first comprehensive attempt to embed AI risk management directly into the federal contracting code, and it could set a precedent for other agencies beyond the Defense Department. If enacted, contractors will need to allocate significant resources to compliance teams, model documentation, and third‑party audits—expenses that could tilt the competitive balance toward larger firms with established governance structures.

The clash also raises questions about the future of AI partnerships between the government and frontier startups. As Anthropic rolls out more advanced versions of Claude, such as the recently announced Opus 4.6, the pressure to demonstrate both performance and safety will intensify. VentureBeat’s recent commentary on “adapting to the new realities of AI” warns that companies ignoring regulatory signals risk being shut out of lucrative federal contracts. For now, the Pentagon’s draft rules remain in draft form, but they signal a clear intent: emerging AI technologies will be subject to the same rigorous oversight that governs traditional defense hardware, reshaping how innovators like Anthropic navigate the federal marketplace.