OpenAI Watch posted an update
European agencies have used several frontier AI models, including OpenAI’s, to examine 27 software projects, while NATO has also given its core institutions access to OpenAI cybersecurity models. The reported work offers concrete examples of AI supporting cyber defence, but it does not mean every vulnerability found can be credited to one model.
Why it mattersWhat happened Help Net Security reports that ENISA and CERT-EU reviewed 27 software projects using several frontier models and tools, including OpenAI’s Codex security plug-in. Maintainers assessed 88 potential issues and confirmed 95% as real. ENISA used OpenAI models in work that found four vulnerabilities in software used across EU institutions; fixes have been released for all four, the report says. The article also says CERT Polska used OpenAI models to help find six vulnerabilities in MikroTik RouterOS. Researchers directed the work and checked findings on real systems; MikroTik released fixes, which CERT Polska says stopped the attacks it had observed.
Discuss: How much evidence should agencies publish about AI’s role in vulnerability discovery before these results count as a meaningful security gain?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.