NATO uses OpenAI Daybreak for cyber defence | Magica
NATO’s OpenAI cyber rollout leaves cost and effectiveness questions open
Editorial Team
••📖6 min read
OpenAI says NATO has used up to 300 billion raw input and output tokens a month for cyber defence. Access through NATO’s technology agency and an AI-agent exercise shows uptake, while costs, comparative performance and NATO-specific security results remain undisclosed.
OpenAI says NATO has used up to 300 billion raw input and output tokens per month for cyber defence.
Crossed Swords introduced Daybreak alongside on-premises AI, with human operators overseeing agents.
European agencies offer examples of verified fixes; NATO’s own security results and procurement costs remain undisclosed.
AI developer OpenAI says it is supplying advanced cybersecurity models to NATO’s core institutions through the military alliance’s Communications and Information Agency. The work brings its Daybreak programme into network protection and military cyber training, while leaving the agency to establish how competing tools compare on cost and operational value.
Crossed Swords 2026 introduced AI-agent training alongside human operators, according to the exercise organiser. Photo: NATO CCDCOE. Source: NATO CCDCOE.
Access is established; NATO outcomes are less clear
OpenAI said in its November 17 announcement that NATO has used up to 300 billion raw input and output tokens a month as part of the work. That is a combined measure of model inputs and outputs, presented as an upper usage figure rather than an average or a sustained monthly rate.
The NATO Communications and Information Agency, or NCIA, already operates the systems this work is intended to protect. NATO’s institutional profile describes an agency created in 2012 by merging existing bodies, providing communications and IT services and protecting NATO networks around the clock through its Cyber Security Centre. A supervisory board representing member countries oversees it, with fundamental financial and policy decisions requiring unanimity. Access for NATO’s core institutions does not establish deployment across every member country’s networks.
Daybreak gives vetted defenders access to advanced AI for authorised security tasks, including software review, investigating suspicious activity, validating vulnerabilities and testing fixes. In its September programme announcement, OpenAI distinguished Daybreak Blue, which uses its mainline models for common defensive work, from Daybreak Red, which gives approved organisations specialised cyber models for more sensitive tasks. It said thousands of defenders across 2,000 approved organisations and workspaces already used Daybreak. The NATO announcement does not specify the access-tier mix behind its usage figure.
New Street Research has initiated Anthropic at Buy ahead of its planned IPO. Its $105 target bets on expanding AI demand, while the filing shows surging revenue and improved customer-serving economics alongside volatile operating results and vast infrastructure commitments.
Editorial Team
OpenAI’s announcement provides no NATO-specific count of confirmed vulnerabilities or deployed patches, contract value, model breakdown or measured reduction in remediation time. Its usage figure measures activity rather than those security outcomes.
The announcement coincided with the opening of NATO Edge 26, NCIA’s industry conference in İzmir, according to a report on the event.
An established exercise adds agents
Daybreak also entered Crossed Swords, the annual cyber exercise organised since 2014 by the NATO Cooperative Cyber Defence Centre of Excellence, or CCDCOE, a cyber training centre in Tallinn. The centre’s exercise account says 2026 introduced a stream in which participants designed, developed, deployed and evaluated AI agents, working alongside them as operators.
The exercise trains both military command elements managing offensive and defensive cyber capabilities and tactical specialists. It also connects cyber operations with special operations training and critical infrastructure protection. The centre described synchronising cyber effects with other operations as a central challenge, with AI involved on both sides. This is a broader operational setting than software patching alone.
Its technical environment used on-premises AI as well as Daybreak, which appeared in the exercise for the first time. More than 20 academic, industry and nongovernmental partners contributed expertise, systems and tools. OpenAI was one participant in that wider environment, rather than its sole technology provider. CCDCOE emphasised maintaining human oversight as agents entered the workflow.
Company-reported summary of the ENISA/CERT-EU review across multiple frontier models. The confirmation rate covers maintainer-assessed issues and does not measure NATO’s results. Source: OpenAI.
Patching evidence comes from other defenders
OpenAI’s announcement gives concrete remediation examples from European agencies outside the NATO deployment. The company says ENISA, the European Union’s cybersecurity agency, used its models to identify four vulnerabilities in software used across EU institutions, with fixes published for all four.
OpenAI also cites a November review by ENISA and CERT-EU covering 27 software projects. Across the frontier models tested, maintainers confirmed 95% of the 88 potential issues they assessed as genuine. That denominator is the issues assessed by maintainers; it does not establish a confirmation rate for every model output. Because the review used several models and tools, including OpenAI models accessed through Daybreak and the Codex security plug-in, the result cannot be assigned to OpenAI alone.
The company further says Poland’s cyber agency CERT Polska used OpenAI models to help identify six vulnerabilities in MikroTik RouterOS, the operating system on MikroTik’s network equipment. Researchers directed the investigation, tested findings on real systems and verified their impact. MikroTik released fixes that CERT Polska confirmed prevented the attacks it had observed. These are concrete examples of human-verified remediation, although they do not measure NATO’s results.
OpenAI says access extends to France’s ANSSI and national cyber agencies in Germany and Poland, while Ukraine’s Ministry of Digital Transformation uses Daybreak for civilian infrastructure defence. The company’s claims are also carried in a report on the rollout.
Procurement needs a comparable measure of value
The NATO work follows a broader push by OpenAI to distribute its cyber capabilities. On September 3, the company announced a $1 billion global commitment covering subsidised Daybreak access, training, technical support and partnerships for frontline defenders. The company targeted consumption of the subsidised access over six months, starting with resource-constrained US defenders and intending to expand to partner countries. That is a programme commitment, not a disclosed NATO contract or evidence that NATO receives a particular subsidy.
There are competing approaches. Rival AI developer Anthropic announced Project Glasswing on April 7 to bring its then-unreleased Claude Mythos Preview model into defensive security work with technology and critical-software organisations. It committed up to $100 million in usage credits and $4 million in donations to open-source security organisations. Its launch terms specified subsequent participant prices of $25 per million input tokens and $125 per million output tokens. An October 6 update says Glasswing members would transition to the specialised tier of its expanded Cyber Verification Program. Those historical terms do not establish NATO’s available options or bill, but illustrate why a combined token count is insufficient for pricing comparisons.
In a statement carried by OpenAI, NCIA praised the models’ potential to accelerate discovery and threat detection while saying it would study consumption, usage patterns and model economics to optimise costs, evaluate competing solutions and guide procurement.
The next evidence needed is a NATO-specific account of verified findings, accepted fixes, threat-detection performance, operator time and total expenditure. OpenAI disclosed no such measures or contract value in this announcement. Comparing those results across tools and deployment environments would help establish whether greater model use translates into better protection at an acceptable cost.
The Alphabet-backed drug developer has announced $7.1 billion in funding in six months. Its new capital will expand proprietary AI models and clinical infrastructure as company-reported preclinical progress awaits testing in patients.
Microsoft enters Ignite with a previously announced Copilot redesign that combines office applications and autonomous agents. Availability, usage-based charges and reliability will determine what that expansion means for customers, while October's missing paid-seat update leaves adoption harder to assess.
A proposed Equity–Pact agreement would give UK film and TV performers consent and payment rights over digital replicas. Members must approve the package, which leaves explicit protections against AI training unresolved.
The French cloud provider plans to release Olala mini 7A1B on November 19, backed by acquired expertise and European public computing support. Its first deployment offers cost claims, but no published benchmarks yet establish its competitive quality.
US stock futures fell before Nvidia's first earnings report since Jensen Huang's departure, with cash generation, tighter export controls and rate-hike expectations clouding the AI trade ahead of Anthropic's expected market debut.
Nvidia has created a standing board committee to oversee national security and export compliance after co-founder Jensen Huang’s departure. Investigation findings and operational changes remain unresolved.
Microsoft's CEO says team sizes, workflows and skills will change as AI reshapes engineering. About 60,000 employees are eligible for voluntary exits, but uptake and the prospect of later layoffs remain unresolved.
Hong Kong officials say financial firms and their human overseers remain responsible for AI agents. With 36 projects selected for sandbox testing, the unresolved task is turning that principle into reliable identity checks, delegated authority and human intervention.
Meta says stronger AI checks and India-specific tuning keep more suggestive and violent content out of Instagram recommendations. Its reporting metric leaves exposure unmeasured, while a government request to change ranking systems drew objections over the ministry’s authority.
Numa’s Series C finances a move from customer conversations into sales and service operations. Automaker distribution gives it reach, but vendor approvals and company-reported savings leave the broader operating-system pitch to prove.
Google’s research tool can turn uploaded sources into interactive dashboards, study aids and trackers. Shared links require full notebook access, personal inputs stay separate, and generation draws on compute-based usage limits.
Records detail Chinese data-center investments linked to Hexi founder Kang Wei, husband of former Megaspeed chief Alice Huang. Filings document financing and Huawei business, but do not establish a route for restricted Nvidia chips.
ElevenLabs plans to acquire Kotoba, the startup behind Japanese speech technology used in Salesforce’s Agentforce Voice. The deal adds a 15-person team and deployed products to its existing Japanese operation, but financial terms and integration plans remain undisclosed.
Tesla plans €3 billion of expansion at its profitable Grünheide factory to build Cybercabs from 2029, eventually targeting up to one million annually. The driverless taxi still lacks German and EU approval, and its capacity target differs from an older plan for the whole site.
Our World in Data’s comparison finds negative views of AI exceed positive ones across five probability-based surveys. The primary results also reveal substantial mixed views, acceptance of some applications and a sharper divide over who controls development and bears its local costs.
NVIDIA's Nemotron for Science pairs an existing open model with nearly 50,000 chemistry questions and scientific post-training recipes. Partner results suggest scope for specialized agents, but the reported accuracy and inference savings do not establish the full cost of adapting them.
AMD says the Lux AI supercomputer is open to scientific users at Oak Ridge. Half its annual capacity supports the Genesis Mission and half is reserved for paid access, including AMD development, while workload limits and undisclosed rates shape its practical value.
Apple is adding app-by-app and folder-specific permissions, repeat approval requests and clearer warnings in macOS 27.3 beta 2. Existing permissions remain intact, limiting the immediate reach of the privacy changes.
OpenAI reportedly discussed buying Goodfire for more than $5 billion, over four times its February financing valuation. No deal resulted. A disclosed paid assessment agreement offers cooperation without ownership, while the technology’s safety benefits remain to be demonstrated.