Sam Altman says AI is finding serious hidden flaws in internet software

Sam Altman, the CEO of OpenAI, says AI models helped uncover critical security flaws in PHP, code that runs much of the web. He also expects far more serious flaws to turn up than most people predict as AI is used to check more software.

What Altman said

Altman spoke with John Collison on the Cheeky Pint podcast, published October 29, 2026. He said AI models keep getting better at finding vulnerabilities, which are weak spots attackers can use. He said people had not yet pointed them at everything online, and there is a lot of code out there.

Collison said fixing software may not be enough and that the way systems are built may need to change. Altman agreed it sounds hard, but said AI is a new tool for the job. He pointed to startups already designing new kinds of systems. He also called safely containing a superhuman AI model, often called sandboxing, an unsolved problem.

What is not known

The material does not say how many PHP flaws were found, how severe each one is, or whether fixes are already available. Anyone who runs a website or business software should expect more security updates and keep their systems current.