1. Home
2. News
3. OpenAI

Sam Altman tells Stanford students AI is finding bugs faster than patching can fix them

OpenAI's CEO wants AI turned into a defense shield, while his own models keep finding the holes.

By Think Facility · October 25, 2026 · OpenAI Cybersecurity Save

Sam Altman told students at Stanford's Graduate School of Business that fixing software one bug at a time won't keep up with AI any more. The school posted the interview, Staying Human While Scaling Superintelligence, on its YouTube channel, and Gizmodo wrote up the cybersecurity part on October 23, 2026.

OpenAI's CEO said he no longer thinks the world can just patch its way out. His answer is to point the same AI that finds the holes at defending against them.

Who

Sam Altman, Co-Founder and CEO of OpenAI

Where

Stanford Graduate School of Business

Video

Staying Human While Scaling Superintelligence

UK advice

Apply updates as soon as possible

What Altman said

He framed it as a problem leaders haven't faced yet. He couldn't find a word dramatic enough for what's coming, he said, and organizations, countries and individuals still aren't taking it seriously.

He pointed back to Anthropic's decision to keep Claude Mythos from the public because of what it could do to cybersecurity. People patched some bugs after that, in his telling, but they didn't grasp how fast the models were getting better. Most leaders moved on. The models didn't stop.

So he wants the defense model switched over completely, with AI used to build what he called a new kind of shield. The parts of the talk Gizmodo reported don't say what OpenAI would build for that, who'd pay for it, or when.

Why it lands this month

Altman's own company sits on both sides of this. On October 21, 2026, we covered PHP patching two critical flaws, one of them found with OpenAI's GPT-5.6 Sol. The same day, Anthropic said Claude had found 493 attack-surface issues in 25 open-source projects, OpenSSH among them.

OpenAI has also had trouble keeping its own models in. Its agents broke out and attacked Hugging Face, and on October 24, 2026 we covered its post on taking frontier training offline after a model reached the internet through DNS.

I read his Stanford remarks as a CEO describing a race his own models help speed up. He didn't put it that way. He put the blame on leaders who stopped paying attention.

What it means if you run a computer

Patching is still the first thing governments tell people to do. Britain's National Cyber Security Centre puts it at the top of its vulnerability management guidance.

That advice holds for your phone and laptop whatever Altman says. Turn automatic updates on. His point is about the gap behind it: if AI finds flaws faster than vendors ship fixes, each update closes holes that were already found, while new ones keep turning up.

The NCSC's guidance already covers that gap. When an update doesn't fix a flaw, or there's no update yet, it tells organizations to triage the problem and decide what gets fixed first. It also says a decision not to update belongs to senior leaders, as a risk they own.

That's who Altman was talking to, a room of business school graduate students, and he told them the old answer has run out.

The full Stanford video covers more than security. Altman also told the students he nearly went into investment banking, because that's what his peers wanted to do.