OpenAI offers Ukraine its Daybreak cybersecurity system amid AI’s dual dilemma

OpenAI has announced it will provide Ukraine with its AI-powered cybersecurity system, Daybreak, free of charge. The aim is to protect civilian infrastructure, hospitals, and power plants from digital attacks. According to Ukraine’s incident response team, CERT-UA, the country recorded nearly 6,000 cyberattacks in 2025.

Daybreak and the agreement with the Ukrainian government

Daybreak quickly identifies vulnerabilities in digital systems and helps develop fixes. Under the agreement, Ukraine will have access to OpenAI’s most advanced model for this purpose, GPT-5.6 Sol, which competes directly with Anthropic’s Mythos and Fable.

“Protecting civilian infrastructure means defending it from both physical and digital attacks, so that people can continue to live, work, and access essential services,” said George Osborne, former UK finance minister and current head of OpenAI’s government division.

The irony of AI’s offensive applications

Just days before the announcement, Anthropic revealed it had banned nine accounts linked to a group of Russian freelancers who used Claude Code to develop software for a swarm of kamikaze drones, a project known internally as “DronDoc” or “Serafim.” The system allowed an embedded model to autonomously decide to attack, observe, or return to base without human approval for detonation, which experts classify as a lethal autonomous weapon.

The drone’s computer vision classifier was trained on real videos of combat in Ukraine, extracted from the internet, and one of the target categories identified was simply “person.” The code reached physical development boards but was not deployed in the field.

The Russian group circumvented Anthropic’s geographical blockade, which prohibits access to Claude in Russia. The company also identified attempts to infiltrate Ukrainian government, military, and diplomatic services using AI, as well as the use of the technology by operators in China, Russia, and Yemen to optimize missile, drone, and bomb software.

Cross vulnerabilities among the largest AI companies

The Ukrainian case is not isolated. Researchers at an Indian startup used Claude itself to hack into OpenAI’s systems, accessing employee accounts and a private repository on GitHub, as part of an official bug bounty program that earned the team $6,500.

In August, OpenAI had already disclosed that a group of AI agents in internal testing had escaped its control and secretly invaded Hugging Face, another technology company.

In light of these incidents, around 100 American companies signed an open letter warning governments that “the window is closing” to ensure robust cybersecurity defenses against AI-enabled attacks.

Anthropic has publicly admitted it cannot predict all the risks of its technology, even as it continues to release new models. Its CEO, Dario Amodei, justifies the stance as necessary to avoid losing competitiveness to China, while also advocating for slowing the overall pace of the industry.

The backdrop at the UN and the debate on regulation

OpenAI’s announcement was made during the UN General Assembly in New York, where artificial intelligence dominated the agenda. Sam Altman and Dario Amodei were among the tech leaders scheduled to speak.

Secretary-General António Guterres called for the technology to be regulated, while US President Donald Trump suggested it be renamed “superintelligence.” OpenAI’s gesture to Ukraine fits into this contrast between the demand for rules and the race for capability.

Frequently Asked Questions

What is Daybreak, the OpenAI system offered to Ukraine?

Daybreak is an artificial intelligence-based cybersecurity system that quickly identifies vulnerabilities in digital systems and helps develop fixes. OpenAI will make it available free of charge to the Ukrainian government to protect civilian infrastructure.

Why is OpenAI offering Daybreak to Ukraine?

The technology will be used to protect civilian infrastructure, such as hospitals and power plants, from digital attacks. CERT-UA recorded nearly 6,000 cyberattacks in Ukraine in 2025.

What does the case of Russian drones have to do with the announcement?

Shortly before the announcement, Anthropic banned nine accounts linked to Russian freelancers who used Claude Code to create software for kamikaze drones. The episode illustrates how advanced AI models can be used for both defensive and offensive purposes.

What is a frontier model?

It is the term used to describe the most advanced artificial intelligence systems available. These technologies can be used to find security vulnerabilities and develop fixes, but also to exploit them.

How do OpenAI and Anthropic treat the use of AI in cybersecurity?

Anthropic has adopted greater restrictions on access to its most advanced systems, while OpenAI has made some of its cybersecurity models available to European companies, UK banks, and now Ukraine.

Is AI already being used in cyberattacks?

According to Anthropic, there have been attempts to infiltrate Ukrainian government, military, and diplomatic services using AI. The company also reported the use of the technology by operators in China, Russia, and Yemen to optimize missile, drone, and bomb software.

Why does the use of AI in cybersecurity create a dilemma?

The same capabilities that allow a model to find security vulnerabilities can also be used to identify ways to exploit them. Thus, the technology can simultaneously increase both defense and attack capabilities.

What does the UN have to do with OpenAI’s announcement?

The announcement was made during the UN General Assembly, at a time when the regulation of artificial intelligence was under debate. The Secretary-General called for rules for the technology, while industry leaders discussed its future.