Skip to content

SecurityOpenAI

Fixed

OpenAI Daybreak: NATO's NCIA runs defender models at up to 300 billion tokens a month, and allies fix AI-found vulnerabilities

Daybreak at a glance

OpenAI published the first dated Daybreak announcement on November 17, 2026: NATO's NCIA uses its defender models at up to 300 billion tokens a month, ENISA found four vulnerabilities in EU-institution software, and CERT Polska found six in MikroTik RouterOS — all fixed.

Ask your AI about this story

Ask what changed in Daybreak and what to check before using it in your business. Identify affected versions, the vendor's mitigation or fix, and how to verify that it is in place.

Sources checked Primary source (opens in a new tab)

Daybreak: status

FixedCritical
Affected
  • MikroTik RouterOS: six vulnerabilities identified by CERT Polska using OpenAI models, including an SSH authentication bypass (CVE-2026-67276) and SSH session privilege manipulation (CVE-2026-86060), both rated CVSS 9.2; the flaws were actively exploited and MikroTik has released fixes, but devices stay exposed until updated
  • Software used across EU institutions: four vulnerabilities identified by ENISA using Daybreak models (CVE-2026-73374, CVE-2026-73405, CVE-2026-73431, CVE-2026-73432), with fixes published for all four
Resolved on
Discovered by
ENISA and CERT Polska researchers using OpenAI models · Found by an AI system

What changed with Daybreak

On November 17, 2026, OpenAI published the first dated announcement for Daybreak, its program that puts frontier AI models in the hands of trusted defenders for authorized security work: reviewing software, investigating suspicious activity, validating vulnerabilities and developing and testing fixes. NATO's Communications and Information Agency (NCIA) gives NATO's core institutions access to the defender models and has used up to 300 billion raw input and output tokens per month. Ukraine's Ministry of Digital Transformation is using Daybreak to strengthen the cyber defence of civilian infrastructure, and OpenAI supported NATO's Crossed Swords 2026 exercise, where participants built, deployed and evaluated AI agents for cyber operations with Daybreak in the training environment.

The results are concrete. The EU's cyber agency, ENISA, used the models to identify four vulnerabilities in software used across EU institutions, and fixes are published for all four. A November joint report from ENISA and CERT-EU describes a review of 27 software projects with frontier models including Daybreak access: of 88 potential issues the models flagged, software maintainers confirmed 95% as genuine. In Poland, CERT Polska used OpenAI models (GPT-5.5-cyber and GPT-5.6-sol, per CERT Polska) to identify six vulnerabilities in MikroTik RouterOS, including two rated CVSS 9.2, and coordinated disclosure with the vendor; MikroTik released fixes that CERT Polska confirmed prevented the attacks it had observed. OpenAI also provides access to national cyber agencies in France (ANSSI), Germany and Poland.

How to fix it: Daybreak

  1. 1. Update

    Update MikroTik RouterOS to a fixed release now: CERT Polska confirmed MikroTik's fixes prevent the attacks it observed, and devices running 7.24.2 or 7.23.4 must update again because the fix for CVE-2026-67278 in those versions was incomplete.

    cert.pl (opens in a new tab)
  2. 2. Update

    Apply the published fixes for the four vulnerabilities ENISA found in software used across EU institutions; fixes are available for all four.

    openai.com (opens in a new tab)
  3. 3. Monitor

    Treat AI-flagged security findings as worth prompt human review: in the ENISA and CERT-EU review of 27 software projects, maintainers confirmed 95% of the 88 issues the models flagged as genuine.

    enisa.europa.eu (opens in a new tab)

Advisories

Daybreak in numbers

Values are copied from the linked source on the date shown. They are not IA Compa measurements.

Raw input and output tokens NATO uses per month, up to (OpenAI)as of · Source for Raw input and output tokens NATO uses per month, up to (OpenAI): openai.com (opens in a new tab)
Up to 300 billion per month
Vulnerabilities ENISA found in EU software, all fixedas of · Source for Vulnerabilities ENISA found in EU software, all fixed: openai.com (opens in a new tab)
4
Vulnerabilities CERT Polska found in MikroTik RouterOS, fixedas of · Source for Vulnerabilities CERT Polska found in MikroTik RouterOS, fixed: cert.pl (opens in a new tab)
6
AI-flagged issues confirmed genuine by maintainers (ENISA/CERT-EU)as of · Source for AI-flagged issues confirmed genuine by maintainers (ENISA/CERT-EU): enisa.europa.eu (opens in a new tab)
95%

Why Daybreak matters for businesses

Why it matters: frontier AI is now doing production security work for governments — finding real vulnerabilities in widely used software and getting them fixed. If your business runs MikroTik RouterOS devices, apply MikroTik's fixes now: CERT Polska found the flaws were already being exploited, and devices stay exposed until they are patched. For everyone else, the pattern is the one to copy: AI-assisted review plus human verification found genuine, fixable flaws at a rate — 95% confirmed in the ENISA review — that makes it worth building into your own security process, whether through a vendor or your own tooling.

Get AI News by email

Verified AI stories, each with its official source, on Monday and Thursday mornings. Add breaking news to hear about the biggest ones as soon as we publish them.

Every story links its official source

Free, in English or Spanish

Unsubscribe with one click, any time

What to send

Sources for this story

  1. Official
  2. Official
  3. Official

Last verified: . IA Compa summarizes public announcements for business readers. We are not affiliated with the companies mentioned. Always confirm details, pricing, and availability with the official source.

  • #openai
  • #daybreak
  • #nato
  • #cyber-defense
  • #cybersecurity
  • #vulnerability-discovery
  • #mikrotik
  • #enisa
  • #cert-polska

Want this working in your business?

We turn new AI capabilities into agents, voice receptionists, and automations you can actually use.