Sam Altman warns AI could uncover far more dangerous security flaws than anyone expects

OpenAI CEO Sam Altman is sounding the alarm about a major cybersecurity problem: AI is getting increasingly good at discovering hidden software vulnerabilities, and we've barely started looking.
Speaking on the Cheeky Pint podcast, Altman warned that the internet may contain far more serious security flaws than even cybersecurity experts anticipate.
Here's the concerning part.
AI models are improving rapidly at identifying vulnerabilities in existing software, but much of the internet's enormous codebase has yet to be systematically examined using these tools.
Recent discoveries involving critical vulnerabilities in PHP highlight how AI-assisted security research can uncover weaknesses that have remained hidden for years.
But this creates a double-edged sword.
The same technology that helps security researchers discover and fix vulnerabilities could also give attackers new capabilities to identify and exploit them.
Altman suggested that patching individual bugs might not be enough. More fundamental changes to software architecture could be necessary to build systems that are resilient against increasingly capable AI.
For businesses, the implications extend far beyond traditional cybersecurity.
Financial platforms, cloud infrastructure, enterprise software, and other critical systems could face growing pressure to adopt AI-powered security monitoring and faster vulnerability remediation.
The bigger question is whether defensive AI can identify and fix security flaws faster than malicious actors can exploit them.
AI isn't necessarily creating all these vulnerabilities. It may simply be revealing how many were already there.
Source: Sam Altman, Cheeky Pint Podcast (October 29, 2026), Computer Weekly