OpenAI’s answer to the coming wave of AI-enabled cyberattacks is to put more advanced AI in defenders’ hands. The company sees a narrowing window to protect essential services, while security concerns persist that money and models alone cannot repair decades of underinvestment.

The alarm had been building before Thursday’s announcement. OpenAI warned organizations last week that they had only months to prepare for AI-assisted attacks, amid concern over attacks on U.S. water systems and scrutiny following the company’s accidental Hugging Face hack.¹ At the same time, its forthcoming Astra model was reported to have reached “critical” cybersecurity capabilities, intensifying the question of whether increasingly powerful systems can be safely controlled.

Sam Altman amplified the company’s pitch in a repost announcing that, alongside GPT-6 Astra, OpenAI would commit “$1 billion to subsidize Daybreak access and frontier capabilities” for frontline defenders of critical infrastructure.Sam Altman (sama)²

At a summit attended by roughly 300 enterprise security leaders and chief information security officers, President Greg Brockman then unveiled Daybreak for Frontline Defenders. The program offers subsidized model access, training, technical support and partnerships for utilities, public agencies and other essential-service operators. Daybreak for America will extend that effort to local governments, water and electricity systems, regional banks and other infrastructure, including a state-and-local training pilot.³

OpenAI says the tools can help smaller organizations inspect legacy code, investigate suspicious activity, validate vulnerabilities and test patches. But the company’s own reality check is stark: many operators still lack people who understand the physical systems behind the digital defenses.³

That unease reaches beyond utilities. Aravind Srinivas argued that increasingly autonomous agents could create GPU capacity and train themselves with little oversight, making “sufficient guardrails and friction” necessary.Aravind Srinivas (AravSrinivas)⁴ The shared diagnosis is urgency; the dispute is whether AI assistance can outrun the structural weaknesses attackers are poised to exploit.