Security1 publisher3 min readPublished
OpenAI pledges over $1 billion in subsidized tokens to Ukraine's infrastructure defenders
The Daybreak agreement puts OpenAI security models into Ukrainian incident response, log analysis and vulnerability validation. OpenAI's national security policy head says the company wants the same deal with other governments.
The Watch · Security desk
What happened
- OpenAI and the Ukrainian government announced a partnership on Wednesday at OpenAI's New York office to put AI tools and subsidized compute behind the defence of Ukrainian critical infrastructure.
- OpenAI pledged over $1 billion in subsidized tokens to the initiative, with access to its cybersecurity models running through the Daybreak program.
- Ukraine's consul general in San Francisco, Dmytro Kushneruk, named the workloads: incident response, threat triaging, login analysis, inventorying systems, code analysis and validating vulnerabilities.
- Kushneruk put Ukraine's volume at roughly 6,000 cyberattacks a year, about 15 a day, after twelve years of Russian cyber operations and physical strikes on electricity and water systems.
Compiled by The WatchSomething wrong?How this is made
Why it matters
- precedent Baker's stated urgency sets up the next round of these deals, and other national CERTs will be asked whether they have one. The vendor writes the token terms, so the reference price for subsidized national defence compute is being set by the seller.
- constraint Subsidized tokens cut the compute bill. They do not shorten the review queue. Kushneruk's framing keeps a defender in the loop on every output, so any real throughput gain is capped by how fast Ukrainian analysts can check what the models hand back.
- exposure Ukraine's triage and vulnerability validation now lean on a US commercial vendor's availability and pricing, and the duration of the subsidy is undisclosed.
- decision US utilities and water operators face the same offer with no wartime justification to settle the regulatory questions for them, and Durakoglu puts regulation among the unresolved points.
Most of the workloads Dmytro Kushneruk named end the same way: a large pile of records becomes a short list a person reads. Threat triaging, login analysis, system inventory, code analysis [4]. Validating vulnerabilities is the one that ends in a decision someone acts on. It is also the only item on that list where a wrong answer either wastes analyst time or leaves a vulnerability unpatched [4].
Kushneruk, Ukraine's consul general in San Francisco, said the point was speed under human review, not substitution. "This is why the object is not to replace the cyber defender with AI, but to make sure the cyber defender acts faster," he said [7]. On incident response he described analysts working through "thousands and thousands of these logs and they have to find what's really important, that's why AI can give capable defenders really much greater advantage and leverage" [6].
OpenAI put a figure on the compute. Over $1 billion in subsidized tokens [3], against Kushneruk's count of roughly 6,000 attacks a year, about 15 a day [9]. Drawn down over twelve months, that is about $167,000 of subsidized inference per attack [19]. CyberScoop's report of Wednesday's announcement at OpenAI's New York office does not state a period for the pledge or a measure the program will be assessed against [20][1].
OpenAI's national security policy head, Sasha Baker, said the company felt "urgency" to reach similar agreements with other governments and industries [17]. She described a period of rushing to get the tools into the hands of critical infrastructure operators, governments around the world, and people who want to patch systems, defend their networks and remediate vulnerabilities. Her reason: "because we know as these tools proliferate out there in the ecosystems, there are going to be bad guys out there that also try to use them" [18]. Greg Brockman signed an open letter earlier this year calling for "collective action" and wide use of AI models to find and fix vulnerabilities before foreign governments and criminals reach the same capability [15]. Politico reported that Sam Altman met US power companies in July about using AI to defend the grid [16].
Naz Durakoglu, minority staff director of the Senate Foreign Relations Committee, said agreement on adopting defensive AI at US critical infrastructure operators runs "pretty much across the board" between the parties [13]. Regulation is still a sticking point [13].
Kushneruk dated the country's preparation to 2015, when Russian GRU hackers shut down the power grid and Ukraine was "maybe not so much prepared" for the fallout [11]. He also cited the 2025 attack on the railway system, after which trains kept running [12]. Twelve years of Russian cyber operations and physical strikes have hit electricity and water [10].
What to watch
- Whether OpenAI names a second government under Daybreak, and on what token terms.
- Whether Ukraine or OpenAI publishes measured triage or vulnerability-validation throughput before and after the tools land.
- Which regulatory sticking points Durakoglu's committee identifies if US critical infrastructure operators are offered a comparable subsidy.