Will All New AI Models Be Classified As Military Grade?

The US government is considering classifying every new advanced AI model as military grade—a move that would fundamentally reshape the global AI industry. If enacted, all frontier AI systems above a specific computing threshold would be subject to strict export controls, classification requirements, and national security oversight.

The proposal, reported by Forbes, follows months of debate in Washington over how to regulate rapidly improving AI capabilities. The core argument: the most advanced models have reached a performance cap that makes them as sensitive as military hardware. Under the plan, any model trained using more than 1e25 floating-point operations (FLOPs) would automatically receive a 'military grade' designation, triggering federal records, restricted access, and potential bans on foreign transfer.

The push stems from growing concern that frontier models—like those capable of advanced code generation, autonomous system design, or biological insight—could be misused by adversaries. Defense and intelligence agencies have urged the White House to treat these models as dual-use technologies akin to advanced semiconductors. The Pentagon’s top AI officer has said that 'without classification, we risk handing our competition the most powerful tools of the century.'

Key figures involved include National Security Adviser Jake Sullivan, who has signaled support for codifying the threshold. Leading AI labs—OpenAI, Google DeepMind, and Anthropic—have privately warned that overly broad classification could stifle innovation and drive talent overseas. Meta’s open-source Llama models would be especially vulnerable, as their weights are currently publicly available.

The plan is still informal, but several bills in Congress already reference 'military grade AI systems' as a category. The bipartisan Artificial Intelligence Export Control Act, introduced in March 2025, would require the Commerce Department to establish a review board for any model exceeding the compute threshold. The board would determine whether the model's release poses a national security risk.

Experts are split. Some argue that without classification, the US risks a security breach akin to the nuclear secrets leaks of the 1940s. Others counter that military grade labels would create bureaucratic bottlenecks and freeze open-source development, ceding leadership to China. 'The term military grade is vague and could cover everything from a chess AI to a medical diagnostic tool,' noted Sarah Chen, a policy analyst at the Center for AI Safety.

The outlook is murky. The administration is expected to issue an executive order later this year requiring all agencies to classify AI systems above the threshold. Industry lobbying will intensify, but the national security argument is powerful. Companies may preemptively self-classify to avoid legal risk. Meanwhile, international allies are watching closely—the EU has proposed its own tiered system, but the US move could set a global precedent.

If all new military grade AI models are locked down, the era of openly shared frontier research could end. Startup founders warn it will 'kill the garage experiment.' But for national security hawks, that is a price worth paying. The coming months will reveal just how far Washington is willing to go—and whether 'military grade' becomes the new standard for every cutting-edge algorithm.