The News, Explained

On November 17, OpenAI said it was working with the NATO Communications and Information Agency, or NCIA, to provide core NATO institutions with access to defensive AI models. NCIA is the NATO agency involved in arranging that access. OpenAI says authorized security teams use its Daybreak program for defensive work such as reviewing software and checking proposed security fixes. Source

OpenAI put NATO’s use at up to 300 billion tokens per month. The figure combines inputs supplied to the models and outputs they produced. A token is a small unit a model uses to process information: a single sentence can contain several tokens. Sending the same document to a model twice, for example, increases input volume. A different measure would be needed to tell whether that led to more completed fixes. Source

OpenAI says it also provided access to the EU Agency for Cybersecurity, or ENISA, and national cyber agencies in France, Germany and Poland. The announcement describes participating organizations and the scale of use. It does not give a NATO-wide count of completed fixes or say how much review times changed. Source

OYOPICK’s Take

Giving public-sector security teams access to AI for review is a meaningful step. To judge its effect, we would like to see the share of issues verified by staff that were actually fixed and the time it took to deploy those fixes alongside usage numbers. Publishing those measures would make it easier to tell whether AI lightens the review workload while improving the quality of defense.