OpenAI has issued one of its strongest warnings yet about the risks posed by rapidly advancing artificial intelligence, cautioning on Wednesday that its next generation of models could present a “high” cybersecurity threat. In a detailed blog post, the maker of ChatGPT said future systems may become capable of generating working zero-day remote exploits against hardened systems or helping orchestrate complex intrusions into enterprise and industrial networks.
The company said the potential for AI-driven offensive capabilities is rising fast, prompting an urgent need for safeguards. OpenAI noted that as its models grow more powerful, they might assist malicious actors in executing real-world cyber operations that traditionally require highly specialised expertise. The concern echoes growing fears among security professionals that AI could accelerate attack development far faster than defensive responses.
To blunt these risks, OpenAI said it is increasing investment in tools that strengthen defensive cybersecurity. These include systems designed to help security teams audit code, identify vulnerabilities and deploy patches more efficiently. The company emphasised that boosting defensive applications is essential as model capabilities expand beyond current containment measures.
OpenAI also detailed the technical controls it is putting in place to constrain misuse. These measures include stricter access controls, hardened infrastructure, outbound-data (“egress”) controls and enhanced monitoring — mechanisms aimed at ensuring that high-capability models cannot be weaponised by unauthorised users.
In a move aimed at building a broader safety ecosystem, the Microsoft-backed firm announced plans to launch a program that would offer qualifying cyberdefense organisations tiered access to more advanced capabilities. This initiative, OpenAI said, is intended to empower defenders without opening the door to malicious exploitation.
To guide long-term safeguards, the company will establish a Frontier Risk Council, bringing veteran cyber defenders and security practitioners into close collaboration with OpenAI’s internal teams. The council will initially focus on cybersecurity but is expected to expand its oversight to other high-risk frontier domains as the technology continues to advance.
