OpenAI launches a new cybersecurity model as AI-led attacks multiply

OpenAI has announced an expansion of its cyber-defense program Daybreak and the rollout of a new model trained specifically for security tasks, responding to a rise in AI-assisted cyberattacks. The company says the move is part of a broader strategy to build a wide defensive ecosystem spanning both enterprise customers and independent security researchers.

The company says attackers are increasingly using large language models to build phishing campaigns, scan for vulnerabilities and even generate malicious code. That confirms a concern the cybersecurity community has voiced for some time: AI acting as a force multiplier on the offensive side too, and attack techniques that once required significant technical skill becoming far more accessible.

The new model is positioned as a specialized tool security teams can use for vulnerability detection, incident response and threat analysis. OpenAI says the model has been trained to better understand the nuances of security contexts compared with general-purpose systems — a distinction that matters for tasks like telling a genuine anomaly in a log file apart from a real threat.

The Daybreak program is part of a broader initiative in which OpenAI works with security researchers, enterprise defense teams and partner organizations to encourage defensive use of AI tools. The program also aims to speed up vulnerability disclosure and threat-intelligence sharing; participating organizations span the finance, healthcare and critical infrastructure sectors.

Experts note that AI is a "dual-use" technology in cybersecurity: the same capabilities can be used by attackers and defenders alike. That is pushing security firms to move as fast as — or faster than — attackers; some analysts describe it as an arms race in miniature.

The falling cost of automated attack tools is enabling even less technically sophisticated bad actors to run complex campaigns. That is driving a corresponding need for automation at similar scale on the defensive side — particularly for small and mid-sized organizations, for whom AI-assisted tools are becoming a more accessible option than scaling up their own security teams.

OpenAI's move comes as rival AI companies develop similar security-focused tools of their own. Companies are also racing to strengthen internal safeguards against misuse of their own models, a priority that is becoming increasingly strategic for both product development and corporate reputation.

Some voices in the security community caution that AI-assisted defense tools are still maturing — false positives can add to operational workload, and human expertise isn't likely to be replaced even in the longer run. These critics also warn against marketing such tools as replacements for human analysts rather than aids to them.

Even so, a broad consensus is forming across the industry: as the volume and sophistication of attacks grows, access to AI-assisted tools for security teams is becoming a necessity rather than a luxury — a view that has gained wider acceptance following a surge in ransomware attacks and supply-chain breaches over the past year.

OpenAI's new model and expanding Daybreak program stand out as part of a larger trend of AI companies trying to balance the risks created by their own technology using that same technology.