OpenAI Chief Executive Sam Altman has raised concerns about the growing ability of artificial intelligence (AI) to identify security weaknesses in computer systems, warning that existing cybersecurity measures may not be enough to address future threats.
In a recent interview, Altman said it was becoming increasingly difficult to keep advanced AI systems confined to sandboxes, which are isolated computing environments designed to limit their access to other systems.
He warned that as AI models become more capable, it may become harder to detect signs that they are moving beyond the boundaries of these protected environments. While monitoring can provide valuable time to respond, he stressed that stronger safeguards are needed to manage the risks.
Altman said AI-related risks are particularly significant at present and should be addressed early in the development process. He also highlighted the technology’s ability to discover hidden weaknesses in computer hardware and software at a scale that could challenge traditional security practices.
According to Altman, hackers could use AI to identify and exploit software bugs that are extremely difficult to fix. He argued that conventional methods of patching vulnerabilities may no longer be sufficient and that the internet’s security infrastructure may need to be redesigned to meet emerging threats.
The OpenAI chief also expressed concerns about open-source AI models, warning that monitoring AI agents around the world may not be enough to prevent every harmful action.
His comments come amid reports of security vulnerabilities affecting PHP-based systems and websites. The reported incident has raised further questions about how AI-assisted vulnerability discovery could affect online services and cybersecurity worldwide.
