Brief IA
›

Daybreak: From Fixed Vulnerabilities to NATO Practices

Models & LLM·Tom Levy·

Daybreak: From Fixed Vulnerabilities to NATO Practices

Daybreak: From Fixed Vulnerabilities to NATO Practices
Key Takeaways
1According to OpenAI, ENISA has identified four vulnerabilities, all of which have been fixed.
2CERT Polska helped identify six flaws in MikroTik RouterOS using OpenAI models.
3NATO has consumed up to 300 billion tokens per month as part of its collaboration with OpenAI.
4Human oversight remains central in the exercises and deployments of AI agents.
Why it matters : CERT Polska confirms that the patches released by MikroTik prevent the attacks observed against the affected equipment.

ENISA and CERT Polska have mobilized OpenAI models in investigations that led to software patches. The company also provides its cyber tools to NATO's central institutions through NCIA. The applications include vulnerability research, human oversight, and training for AI agents.

ENISA identifies four vulnerabilities, CERT Polska finds six

According to OpenAI, the European cybersecurity agency ENISA has identified, using its models, four vulnerabilities affecting software used in European institutions. All have been patched. In Poland, CERT Polska relied on models from the same provider to help discover six vulnerabilities in MikroTik RouterOS, then organized their disclosure with the publisher. MikroTik delivered patches that CERT Polska confirmed blocked the observed attacks and allowed operators to secure the affected devices.

A broader review, presented in November by ENISA with CERT-EU, covered 27 software projects. It utilized several advanced models, including those available in Daybreak, as well as tools like the Codex security plug-in. According to these organizations, maintainers confirmed the reality of 95% of the 88 potential issues they assessed among the tested models.

Crossed Swords 2026 trains operators to supervise agents

At CERT Polska, researchers retained control of the investigations: they tested the results on real systems and verified the effects of the vulnerabilities. OpenAI attributes to AI an acceleration of the analysis and exploration of potential vulnerabilities. The company also sees the validations obtained by ENISA and CERT-EU as evidence of the value of a partnership between human expertise and AI.

Supervision was also on the agenda of Crossed Swords 2026, organized by the CCDCOE, NATO's Cooperative Cyber Defence Centre of Excellence, with the support of OpenAI. A new training program offered participants the opportunity to design AI agents, deploy them, and then evaluate their behavior for cyber operations. They used Daybreak in a technical environment, taking on the role of operators responsible for maintaining human oversight.

OpenAI equips NCIA and European administrations

OpenAI is working with NCIA, NATO's Communications and Information Agency, to make its advanced cyber models accessible to the Alliance's central institutions, aiming to help protect allied networks. The company describes Daybreak as a tool reserved for trusted defenders, within an authorized framework: code review, investigation of suspicious activities, validation of vulnerabilities, or preparation of remediations. It also states that it has opened its models to ENISA and national agencies in France, Germany, and Poland.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

In Ukraine, the Ministry of Digital Transformation is using Daybreak for the defense of civilian infrastructure. The tool is intended to help teams identify software vulnerabilities more quickly and then develop and test patches, as the essential services they protect face persistent cyberattacks. These applications concern systems on which populations, administrations, and economies depend.

Sasha Baker, National Security Policy Lead at OpenAI, advocates for access to the most advanced technologies for those protecting democracies. She presents Daybreak as a support for NATO teams responsible for networks essential to the Alliance, enabling them to fix vulnerabilities before they can be exploited. She aims to deepen this cooperation with NATO and its allies. More broadly, OpenAI plans to continue its government and international partnerships to strengthen system resilience, accelerate the handling of weaknesses, and protect everyday services.

NCIA wants to clarify the costs of the models

NATO has consumed up to 300 billion tokens monthly, including inputs and outputs, in this collaboration, OpenAI reports. Dylan Browne, General Manager of NCIA, anticipates that broader usage will provide better insights into this consumption, user practices, and the economic parameters of the models. This knowledge should, according to him, help control costs, compare competitors, and guide purchasing decisions.

Dylan Browne describes these models as a generational advancement: he attributes to them an accelerated discovery of vulnerabilities, detection of emerging threats at machine speed, and an advantage for defenders. He also believes they enhance cyber resilience and operational advantage for the Alliance, while increasing its capacity to protect critical digital infrastructures at an unprecedented scale and speed. OpenAI asserts that initial uses support the security and freedom mission of member states, develop the skills of multinational teams ensuring continuous monitoring, and produce valuable lessons in Europe and within the Alliance. The company reaffirms its commitment to NATO principles regarding responsible AI adoption and its dedication to providing safe tools.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.