The Timeline So Far

Since OpenAI signed its Pentagon deal in late February 2026 — hours after Defense Secretary Pete Hegseth declared Anthropic a "Supply-Chain Risk to National Security" — Sam Altman has been in an increasingly uncomfortable spotlight on Capitol Hill.

The pressure has been building for months, and lawmakers are not letting it go.

What Happened on Capitol Hill

Altman met with a select group of lawmakers in Washington, D.C. in a session that sources described to CNBC as loaded with "serious questions" about OpenAI's approach to warfare and its Pentagon partnership.

Sen. Mark Kelly, D-Ariz., told CNBC's Emily Wilkins that the group talked "in detail" about surveillance and how AI could be used within a kill chain. His words, not ours.

Kelly called it a "good discussion" but was direct: "There's got to be guardrails in place, and we've got to make sure that we're always thinking about the Constitution and making sure that we comply with it."

According to The Hill, Altman also tried to distance himself from the AI industry's broader lobbying push, even as millions of dollars flow into Washington from tech sector interests. Whether lawmakers bought that distancing act remains unclear.

The Anthropic Blacklist Backstory

Anthropic was not just denied a contract. It was blacklisted by Hegseth as a national security supply-chain risk.

Why? According to CNBC, Anthropic had been trying to renegotiate its DOD contract but talks collapsed because the Pentagon wanted "unfettered access" to its models for all lawful purposes. Anthropic wanted explicit assurances its tech wouldn't be used for fully autonomous weapons or domestic mass surveillance.

The DOD said no. Anthropic walked. Hegseth dropped the hammer.

Within hours, OpenAI stepped in and signed a deal. Altman posted on X that the DOD agreed to prohibit domestic mass surveillance and autonomous weapons use — two things he called OpenAI's "most important safety principles."

OpenAI published a contract excerpt. It says the DOD "may use the AI System for all lawful purposes." The company says its "safety stack," contract language, and existing laws will prevent abuse. That's a lot of trust to place in bureaucratic good faith.

The Policy Shift That Started All This

None of this Pentagon business would have been possible without a quiet but consequential decision OpenAI made in January 2026: it removed language from its usage policies that explicitly prohibited military applications.

According to techbuzz.ai, that policy revision unlocked DOD contracts but created direct tension with OpenAI's founding mission. The company once positioned itself as the safety-first alternative to reckless AI development. Now it's embedded in national security infrastructure.

Critics — including former OpenAI employees who went public — questioned whether this constitutes mission creep. The question deserves scrutiny.

Trump's AI Executive Order: The Good and the Watch-Outs

Separately, Trump signed an executive order titled "Promoting Advanced Artificial Intelligence Innovation and Security." The core focus is cybersecurity for federal systems — patching vulnerabilities, directing the NSA and the Cybersecurity and Infrastructure Security Agency to build a classified benchmarking process for frontier AI models.

The order also includes a voluntary program for AI developers to share new models with the government for assessment.

According to Reason, the order explicitly states it "shall not be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement" for AI development or release.

Jessica Melugin of the Competitive Enterprise Institute said it "wisely stops short of calling for mandatory government licensing" but warned it "leaves plenty of room for future regulatory overreach."

Cato Institute analyst Juan Londoño praised the voluntary commitment but flagged a real concern: the lack of clear criteria for what counts as a "covered frontier model" and the government's role in picking "trusted partners" gives the executive branch significant discretion. That discretion could be used against companies that find themselves in conflict with the administration. Today's voluntary program is one executive order away from becoming tomorrow's licensing regime.

Autonomous Weapons: The NDAA Fight Coming

AI policy groups are now urging leaders on the House and Senate Armed Services Committees to add autonomous weapons guardrails directly into the National Defense Authorization Act, according to The Hill.

This fight is separate from the OpenAI-DOD deal but directly related. If Congress doesn't draw hard lines in the NDAA, there are no hard lines. Contracts with safety language are only as good as the administration enforcing them.

The Structural Problem

Left-leaning outlets like CNBC focus heavily on Altman's reassurances and quote Kelly's concerns without pressing on the structural problem: voluntary safety commitments from a private company are not law. Right-leaning coverage emphasizes the Trump order's restraint without asking who gets to define "trusted partners."

Both miss the core issue: the U.S. government is now deeply dependent on private AI companies for national security, and neither Congress nor the executive branch has codified enforceable rules for what those companies can and cannot do with that access.

The Bottom Line

AI systems are being woven into military kill chains right now. The guardrails are a company blog post, a contract excerpt, and a senator saying he asked serious questions. This is not a regulatory framework. If Congress doesn't act through the NDAA, the rules will be whatever OpenAI and the Pentagon agree to behind closed doors.