OpenAI warns of rising cybersecurity risks from next-gen AI models

With future models edging toward capabilities used by advanced hacking groups, OpenAI is introducing stricter safety layers, selective access to powerful tools and a dedicated advisory council to reinforce cybersecurity and prevent misuse.

OpenAI has issued one of its strongest warnings yet about the cybersecurity implications of its next generation of AI models, cautioning that rapid capability gains could soon place advanced offensive techniques within reach of automated systems. The alert comes at a time when the company is under mounting competitive pressure to outpace Google’s Gemini project and deliver more powerful iterations of ChatGPT.

In a detailed blog post, OpenAI said its future AI models may approach tasks historically handled only by sophisticated hacking groups, such as generating working zero-day exploits or contributing to high-level intrusion plans targeting industrial networks. The company acknowledged that the pace of progress could soon cross into territory that raises “high” cybersecurity risks.

Company prioritises defensive capabilities over offensive potential

Despite the severity of the concerns, OpenAI emphasised that its goal is to ensure AI advances serve cybersecurity defenders—not enable attackers. The company said it is “investing in strengthening models for defensive cybersecurity tasks and creating tools that enable defenders to more easily perform workflows such as auditing code and patching vulnerabilities.”

As part of this shift, OpenAI plans to embed layered safeguards into its next-gen systems, including tighter access controls, hardened internal infrastructure, egress monitoring and expanded technical oversight. The company is also preparing a tiered access program that will grant vetted researchers and cyber defence teams early access to enhanced capabilities under strict qualification standards.

New advisory council to oversee high-risk AI domains

To bolster oversight, OpenAI is forming the Frontier Risk Council—a new advisory group comprising seasoned cyber defenders and security specialists. Initially focused on cybersecurity, the council is expected to expand its remit to other high-risk domains associated with frontier-level AI models.

Internal push to accelerate training raises questions

While OpenAI’s public commentary centres on responsible scaling, internal developments reflect a growing urgency. Reports indicate that CEO Sam Altman has encouraged teams to accelerate progress by relying more heavily on user-generated data, including one-click feedback from ChatGPT interactions, rather than depending primarily on expert evaluators. Supporters say this approach accelerates model learning, but critics warn it may introduce inconsistencies and reduce oversight.

As OpenAI tightens safety measures while simultaneously speeding up development, the company faces a delicate balancing act: pushing innovation forward without compromising global cybersecurity.