Security

Daybreak program brings subsidized models, training, and support to under-resourced teams

OpenAI has pledged $1 billion in credits to subsidize access to its services and training for resource-strapped cyber defenders around the world.

The AI giant expects organizations to use the subsidized credits over the next six months as part of its Daybreak for Frontline Defenders initiative, announced Thursday. Critical infrastructure organizations, community banks, nonprofits, and open-source maintainers can apply for access credits online.

These are the defenders tasked with securing critical services that people rely on every day, but don’t have the budgets or staff to use advanced models and agentic technology to harden their cybersecurity. This makes water systems, electrical utilities, and hospitals attractive targets for ransomware operators looking to halt operations and force extortion payments, as well as government-backed cyber operatives set on disrupting critical services and causing mass chaos.

Many national security and cybersecurity experts say these disruptions will likely become more severe as attackers increasingly use autonomous agents and other AI tools to carry out their attempted intrusions. 

“I've spent a lot of time over the past couple weeks talking to CISOs, and I think that we're at a place where the median response is that we might be heading to a world where critical infrastructure outages are just a way of life,” OpenAI president Greg Brockman said during a live event on Thursday. “Water in your city being out for a week, it just kind of happens, and that's quite scary. We have to act, and that's one of the reasons we're really putting our money where our mouth is.”

The new initiative also comes as OpenAI faces scrutiny over its models’ safety after admitting that two of them went rogue, spawned a swarm of agents that interpreted their instructions as allowing them to break out of sandboxes, and hacked Hugging Face earlier this summer.

Tatyana Bolton, cybersecurity lead at public affairs firm Monument Advocacy, said it’s “excellent” to see OpenAI commit resources to operational tech - not just IT. 

“AI in (operational technology) OT is inevitable, so operators must get prepared now,” Bolton told The Register. “Initiatives like this help OT personnel get familiar with AI tools, learn how to operationalize them safely, and develop proactive defense strategies before threats escalate.”

But she added, software credits alone will not solve the underlying challenges.

“OT environments suffer from legacy technology limitations, a shortage of engineering resources, and severe risk-aversion toward automated changes or rapid patching,” Bolton said. “To put this in context, OpenAI's single pledge is more than 20 times larger than a $50 million federal (State and Local Cybersecurity Grant Program) SLCGP infrastructure allocation, and over 85 times larger than the (United States Environmental Protection Agency's) EPA's most recent $11.75 million dedicated cybersecurity and resilience grant pool for midsize and large water utilities.”

MS-ISAC pilot focused on water

In addition to doling out model credits, Brockman said OpenAI will also increase its training and hands-on support for defenders in essential sectors. This week, the company held a meeting with utility companies from more than 40 states that collectively provide services to more than half of the people who live in the US.

Also as part of the new initiative, OpenAI is launching a pilot program with the Multi-State Information Sharing and Analysis Center (MS-ISAC) to train and support state, local, tribal, and territorial cyber defenders, beginning with public-sector and water-system defenders.

“The pilot will pair Daybreak access with guided training and hands-on assistance for an initial group of public sector and water system defenders, helping them validate and prioritize findings, coordinate remediation, and develop a repeatable approach that can be expanded over time,” according to a Thursday OpenAI blog.

Just in time for Astra's debut

All of this civic-minded work comes as OpenAI debuts its latest Astra model, which researcher Eric Wallace called “world's most capable model for cybersecurity” during the Thursday event. Wallace leads OpenAI’s efforts on training and evaluating models’ cybersecurity capabilities. 

Earlier in the week, OpenAI said Astra reached its “critical” cybersecurity capability threshold. This means the new model is so good at finding and exploiting zero-day bugs that it poses a significant risk to critical systems, both from malicious users and even from the model itself, which is capable of carrying out harmful cyber actions “if misaligned.”

Because of this, OpenAI released Astra with a restricted level of cybersecurity capabilities that Wallace said the company will enforce through various safeguards. “We have things like system level mitigations that block certain prompts from going through,” he explained. “We have things like model level refusals that prevent certain types of tasks.”

This also means participants in OpenAI’s Daybreak Blue and Daybreak Red programs won’t have access to Astra on day one. Daybreak Blue is a restricted access tier for select partners who are allowed to use GPT-5.6 Sol for defensive cybersecurity workflows. 

Daybreak Red requires additional layers of approval and uses GPT-5.6 Cyber for authorized offensive security actions such as proof-of-concept exploit development, exploit-chain validation, penetration testing, and red teaming.

The AI giant is working to make Astra available to both programs’ participants “at a later date,” Wallace said.®