Microsoft CEO Satya Nadella says smarter AI does not mean safer AI

Satya Nadella, the head of Microsoft, which makes Windows and the Office suite, just admitted he got a major bet wrong. On a new episode of the 20VC podcast, host Harry Stebbings asked the tech chief what he had rethought over the past year.

Nadella confessed he used to believe that as artificial intelligence systems grew more capable, alignment (making sure the technology respects human values) would happen all on its own. Instead, he watched computer power scale up just as he predicted, while keeping the models safe proved to be an entirely separate and much tougher mountain to climb.

"I used to always think that alignment would just fall out," Nadella said on the show, according to Business Insider. "I no longer believe that. I think this alignment thing may be a harder, different problem."

The shift in thinking comes as the software giant tightens its own defenses. Nadella recently warned that companies must treat advanced artificial intelligence models like insider threats and assume they are already compromised. He called for rigid human controls and new industry standards to box in unpredictable software.

He is not alone in sounding the alarm. OpenAI Chief Executive Sam Altman and Anthropic Chief Executive Dario Amodei have both raised similar red flags, while a July letter signed by more than 1,000 tech workers warned that capability is outpacing control.

For any business owner watching the massive spending on software tools, the message is clear: the people building the technology admit they are still figuring out how to keep it on a leash.

Photo: Brian Smale / Microsoft, CC BY-SA 4.0, cropped