NSA Seeks Access to ‘All’ AI Models as Spy Agency Takes Key Role in Voluntary Testing Program

Spy agency signals sweeping interest in examining frontier systems

The National Security Agency wants to inspect “all” artificial intelligence models, a top official said, revealing an expansive ambition as the spy service takes a leading role in the US government’s new voluntary AI testing initiative.

Speaking to Nextgov/FCW, the NSA’s deputy director did not identify which companies are participating or whether any model developers have already agreed to open their systems to scrutiny. The remarks nonetheless underscore how Washington is treating the most advanced AI not as a mere commercial or consumer technology, but as a national security priority.

“We want access to all AI models,” the senior NSA official said, according to the report.

The agency’s involvement in the testing program places it at the centre of federal efforts to evaluate frontier models before they are widely deployed. The voluntary framework is still taking shape, and the NSA’s outsize role raises questions about how the evaluations will function in practice: what the scope will be, how sensitive technical details will be protected, and whether companies will willingly hand over deep access to their proprietary systems to the US intelligence community.

From voluntary testing to national security oversight

The programme relies on companies choosing to cooperate, rather than a legal mandate—at least for now. The NSA’s desire to examine every major model suggests it sees the voluntary mechanism as an opportunity to create a de facto standard. Big tech firms that develop frontier models may face pressure to participate if staying outside the tent is interpreted as a refusal to engage with security authorities.

Because the official disclosed no names, it remains unclear whether OpenAI, Google, Anthropic, Meta or other leading AI developers are already in talks with the NSA, or if any have declined. Civil‑liberties groups and AI ethicists are likely to flag concerns that a spy agency, whose primary mission is signals intelligence, could use model testing as a backdoor to demand data or embed surveillance capabilities.

Broader context: the AI policy landscape

The NSA’s move comes against a backdrop of accelerating federal AI action. The White House has published a Blueprint for an AI Bill of Rights and secured voluntary safety commitments from leading developers. The National Institute of Standards and Technology has been building evaluation frameworks, while other agencies explore their own regulatory angles. The NSA’s role injects a distinct intelligence‑gathering perspective into that mix.

For the moment, the testing programme remains shrouded in secrecy. Details about what exactly the agency will look at—bias, robustness, potential for misuse, underlying training data—are not public. Neither is the legal basis for how the NSA would handle any model code or data it obtains. These unanswered questions are likely to intensify scrutiny as the programme moves forward.

What to watch next

• Which companies publicly confirm participation—or choose to remain silent.
• Reactions from civil‑liberties advocates and privacy watchdogs concerned about intelligence agency overreach.
• Whether Congress or other regulators push to convert the voluntary scheme into mandatory testing, giving the NSA a statutory foot in the door.
• Statements from other federal bodies such as the Department of Commerce or the White House OSTP explaining how the NSA’s role fits into the wider AI safety architecture.

For now, the NSA’s plain statement that it wants access to “all” AI models sends an unmistakable signal: in Washington’s eyes, frontier AI is a domain where national security interests are non‑negotiable—and the spy agency intends to be at the table.