OpenAI just became a defence contractor for NATO

A glowing AI shield emblem over a military command bunker with network lines

A few years ago OpenAI was a research lab that made a chatbot. On 17 November it announced it is working with NATO’s Communications and Information Agency to give the alliance’s core institutions access to advanced cybersecurity models, to help protect allied networks. Read that plainly: the company behind ChatGPT is now, in effect, a defence contractor for the Western military alliance. And it is not alone.

What the deal is

The arrangement gives NATO’s central bodies access to OpenAI’s frontier models for cyber defence, the work of spotting intrusions, hardening systems and responding to attacks on the networks that hold the alliance together. The details are thin, as these announcements tend to be, but the direction is unmistakable. The timing makes it sharper still: it lands in the same week Google reported the first cyberattack run entirely by an AI with no human in the loop, and the FBI warned that offensive AI tools have been used against organisations worldwide since the summer. NATO is arming its defences at the exact moment the threat is going autonomous.

The bigger shift

This is one move in a much larger pivot. The frontier AI labs started out selling productivity and promising to benefit humanity. They are now racing toward national security, because that is where the deepest budgets and the greatest urgency live. OpenAI is working with NATO. Other labs have their own government and defence engagements. The sector that spent its early years talking about safety and alignment is quietly becoming part of the military-industrial base, and the commercial story and the war story are merging into one.

The case for, and the unease

There is a genuine case for it. The threats are real, as this week’s autonomous-attack news makes clear, and if anyone can build systems fast enough to defend against machine-speed attacks, it is the companies building the machines. Allied networks protected by frontier models is not a dystopian idea on its face. It may simply be necessary.

The unease is about concentration. The same handful of companies building the most capable models on earth, models that can just as easily be pointed at offence, are now also the ones defending the West’s institutions. That is an enormous amount of power and insight gathered into very few hands, with oversight that has not remotely kept pace, and a governance model that still amounts to trust us. When the firm that builds the sharpest sword is also hired to hold the shield, the obvious old question returns: who is watching the people doing the watching.

What this means

AI’s two stories, the commercial one and the military one, are collapsing into each other, and OpenAI arming NATO is a milestone in that collapse. It is probably sensible defence policy in a week that proved the threat is real. It is also a reminder that the companies we are handing our security to are accountable, for now, mostly to themselves. Useful and uncomfortable are not opposites here. They are the same deal.

Edgar Friendly, calling the AI bullsh*t. More tools, more swearing: toptoolstack.com.

Did you know? The organisation taking OpenAI’s models is NATO’s Communications and Information Agency, the body responsible for the alliance’s technology backbone. The networks it protects coordinate the defence of more than 30 countries.

Sources: Anadolu Agency; The Hill.

Get the free weekly stack: the AI tools and moves that matter, hype filtered out.Subscribe free →
Scroll to Top