Lawmaker Says Spy Agencies Need Early AI Model Access

Rep. Jim Himes, the top Democrat on the House Intelligence Committee, delivered blunt testimony at Politico's Security Summit: withholding advanced artificial intelligence models from U.S. spy agencies would be "insane."

The Connecticut Democrat's remarks cut through the usual diplomatic language surrounding government technology access. "Making sure that, in particular, where our real computational brains are, the National Security Agency, making sure they have access to the most capable hacking tools … it would be insane not to do that, right?" he said during Tuesday's panel discussion.

His comments arrive as the Trump administration reportedly debates whether the Commerce Department or the intelligence community should oversee AI model evaluations. Commerce officials are pushing back against a White House proposal to house an AI evaluation center within the intelligence community, according to Nextgov/FCW reporting.

The stakes involve real, classified technology. The NSA has been testing Mythos, a major Anthropic model held back from full public release due to its substantial cyber capabilities. Multiple people familiar with the matter confirmed the spy agency's access to this restricted model.

Himes acknowledged the Commerce Department should also play a role in AI policy. "Across the government, we should be looking at these capabilities," he said. He added that the U.S. ought to be cultivating — not damaging — its relationship with technology producers.

This nod references the ongoing legal complaints Anthropic has lodged at the Defense Department. The Pentagon deemed the company a supply chain risk earlier this year after Anthropic refused to meet certain demands. Himes doesn't think this spat has set back the intelligence community in the near term, though he warned: "if this drags out, if [Defense Secretary] Pete Hegseth gets a bee in his bonnet about this and just decides to target because his ego is damaged … that will be a massive liability for United States national security."

Officials are circulating draft policy documents with language clarifying the government's ability to use private sector tech without outside stipulations. It's not clear if the contracting language is part of a coming executive order or a separate policy initiative (another bureaucratic black box, honestly).

The physical reality of this access matters. When intelligence analysts work with frontier models, they're not clicking through consumer interfaces. They're operating in classified environments where latency, data handling, and model behavior under adversarial conditions determine whether a cyber threat gets caught or slips through. The difference between Mythos being available to the NSA versus the public isn't just a matter of timing — it's about who gets to weaponize the vulnerabilities first.

President Donald Trump is making a planned trip to China this week, where he is expected to discuss AI matters with Chinese President Xi Jinping. The timing underscores how domestic AI policy debates intersect with international competition.

The ongoing discussions highlight how the Trump administration is closely examining cyber threats brought on by advanced AI models and is looking to take a more hands-on approach toward the AI sector, despite prior laissez-faire positions.

Whether this early access framework actually improves national security or just creates another layer of government-privileged technology remains the real question. The public gets to wait in line while the agencies get the VIP pass — a tradeoff that feels increasingly familiar in an era of classified AI development.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn