OpenAI and NATO have partnered on cyber defense, it was announced on the first day of the "NATO Frontier 26" conference in Türkiye.

The specific one used is the NATO Communications and Information Agency. This agency is NATO's technical and cyber hub, with more than 3,000 civilian and military experts. OpenAI will provide its network security model to investigate suspicious activity, discover vulnerabilities, and deploy fixes.

There is an astonishing number: NATO uses 300 billion tokens every month in network security work.

This is just now.

The general manager of NATO said that the model can detect new threats and take countermeasures at "machine speed," making vulnerability discovery much faster. Listening to this week's context, the meaning is very different: Bank of America has been down in the past few days because AI has dug out too many vulnerabilities and patches have been rushed too quickly; on the other hand, a researcher used Codex to dig out high-risk vulnerabilities in Java in one hour. Both the attacker and the defender have the same type of models, so whoever finds them first and who makes up for them first becomes a matter of speed.

For OpenAI, this is also an image repair. In the past two months, it has been closely watched over safety issues. It was sued in California and filed in Arizona. Its own model also had an accident when it ran out of the sandbox. Giving the security model to a military alliance to defend the network is tantamount to saying: Our model can also be on the defensive side.

As for whether the model will be used offensively, the announcement did not mention it.