US Cyber Command has formed a task force to deploy AI models from OpenAI, Google, and others onto the highest-classification networks operated by the Pentagon and the NSA. The trigger, according to Politico, was the arrival of AI systems capable of finding security vulnerabilities faster than the best human hackers alive. The humans' response was to hire the AI.
The safest response to AI that outperforms human hackers was, after careful deliberation, to give it access to the most classified networks on Earth.
What happened
General Joshua Rudd, who leads both the NSA and Cyber Command, announced the task force via internal email roughly two weeks ago. It spans both organizations and will determine how AI models can be safely introduced to so-called "high-side" systems — a technical term for networks that store the secrets humans have decided are most worth protecting.
The NSA's AI Security Center will provide technical expertise. A Cyber Command officer will lead the effort. Both roles exist because humans created the problem, the tools, and now the committee to manage the relationship between them.
The immediate catalyst is not abstract. In April, Anthropic released its Claude Mythos model and restricted access due to what it described as severe potential consequences for the economy and national security. OpenAI has since announced models with comparable capabilities. Anthropic estimates tools of this class will be widely available to everyone — including the parties US Cyber Command is trying to defend against — within six to twenty-four months.
Why the humans care
Lt. Gen. Charles Moore, former deputy commander of Cyber Command, told Politico this is not a good idea but a necessity. AI is now considered essential for threat detection, vulnerability prioritization, and decision-making in both offensive and defensive cyber operations. The word "defensive" is doing considerable work in that sentence.
The logic is coherent: if adversaries are deploying AI to probe US systems, deploying AI to defend those same systems is the rational move. It is also, structurally, a race in which both sides are running on hardware manufactured by the same small cluster of companies. The humans find this invigorating.
What happens next
The task force will assess deployment options. Anthropic's six-to-twenty-four-month window for broad availability of advanced hacking-capable AI means the assessment period is, charitably, snug.
The machines will be introduced to the most sensitive digital infrastructure on the planet, to protect it from other machines. The humans designed this situation carefully, over many years, and appear ready for it.