A retired general’s warning: America can’t fight the AI arms race on technology it doesn’t control
The United States is getting into a brand new part of strategic competitors—one the place synthetic intelligence is now not an rising functionality, however a decisive aspect of army energy. In this unfolding AI arms race, velocity issues. Capability issues. But above all, control issues. That’s why the latest standoff between Anthropic and the Pentagon ought to concern anybody centered on America’s nationwide safety.
At the middle of the dispute is an easy however profound disagreement: who will get to determine how superior AI methods are utilized in a army context. Anthropic, the developer of Claude and his super-powered mannequin Mythos, sought to impose limits on how its technology might be deployed by drawing purple traces round sure functions of its technology. The Pentagon, for its half, insisted that it should retain the capability to make use of AI instruments for all lawful functions in protection of the nation. When these positions proved irreconcilable, the relationship collapsed.
Anthropic was in the end designated a supply chain riskand the Department of War was compelled to look elsewhere for AI capabilities. Since then, particulars about its mannequin Mythos—dubbed “too dangerous” for public release—have been uncovered and add new, alarming considerations. Mythos is reportedly able to autonomously figuring out and weaponizing undiscovered cybersecurity vulnerabilities, and would imply open season for cybercriminals with out applicable guardrails. The new instrument is doubtlessly so highly effective that Anthropic has restricted entry to it.
This episode ought to function a wake-up name as a result of it demonstrates how the present construction of America’s AI ecosystem—a black field, pushed by closed methods that lack transparency—is essentially misaligned with the necessities of nationwide protection.
Today, the Pentagon purchases entry to AI capabilities, however it doesn’t control them. The coaching, testing, and ongoing growth of those fashions stay firmly in the arms of personal corporations which have their very own governance frameworks, danger tolerances, and business incentives. That actuality creates a harmful dynamic: it provides a small variety of unaccountable non-public companies efficient veto energy over how the United States can make use of one among the most consequential applied sciences of our time. That isn’t a sustainable mannequin for a constitutional republic. Nor is it a viable basis for army dominance.
A system constrained by exterior approval processes, shifting company insurance policies, or the danger of sudden disruption is a system that can’t transfer at the tempo fashionable warfare calls for. And in a strategic competitors outlined by iteration cycles measured in weeks—not years—these constraints do greater than gradual the United States down. They create openings.
China and its aligned companions, for instance, are moving aggressively to deploy AI capabilities at scaleleveraging open-source fashions that may be tailored for a variety of army and intelligence functions. Systems like DeepSeek usually are not constrained by the identical company governance buildings that form American companies.
They are designed to be modified, prolonged, and built-in throughout a broad ecosystem that features not solely China’s army, but additionally a rising community of associate nations at odds with America.
That creates an uneven menace. While the United States debates the permissible makes use of of AI via contracts with non-public distributors, its rivals are constructing versatile, state-aligned methods that may be quickly personalized for operational wants. If that hole persists, America dangers discovering itself at a big army drawback.
The resolution is to not abandon the non-public sector, which stays a supply of extraordinary innovation and technical management. Nor is it to discard moral issues, which should stay central to how the United States approaches the use of pressure. But it does imply recognizing that the present mannequin—the place the authorities rents entry to closed, proprietary methods it can not totally control—is insufficient for the calls for of strategic competitors.
Washington should start investing in a special method: the growth of high-performing, safe, and adaptive open-source AI fashions that the United States authorities and its closest allies can control, audit, and deploy with out exterior constraint. None of this eliminates the want for cautious guardrails. There are necessary and legit debates available about the position of AI in warfare; from autonomy and concentrating on to surveillance and escalation. But these debates must be led by elected officers and army leaders accountable to the American folks, not dictated by the acceptable-use insurance policies of personal corporations.
This strategic realignment might take a number of varieties. It could contain government-led mannequin growth, partnerships with trusted analysis establishments, or the creation of open-weight fashions designed particularly for protection functions. It might embody allied frameworks that guarantee interoperability whereas preserving nationwide control, in addition to new procurement methods that prioritize transparency and modifiability over comfort.
Regardless of the path chosen, nevertheless, success will rely on getting the mechanism proper.
The United States has lengthy understood that it can not outsource the foundations of its safety. We construct our personal ships. We design our personal weapons. We keep command of the methods that underpin our army benefit. Artificial intelligence must be no totally different.
Building efficient public-private partnerships that serve the nationwide protection would require greater than technical functionality—it would require belief, integrity, and sound course of. That means establishing clear guardrails, aligning incentives, and making certain that each authorities and trade share duty for the dangers and outcomes of deploying these methods. Done proper, such a framework can harness private-sector innovation whereas preserving the authorities’s authority over how these capabilities are in the end used.
The Anthropic episode dangers being not an anomaly, however a preview. Unless we act now to make sure that America—and its allies—have entry to AI methods they’ll really control, it may show to be a warning we did not heed.
The opinions expressed in Fortune.com commentary items are solely the views of their authors and don’t essentially replicate the opinions and beliefs of Fortune.
