Microsoft CEO Nadella says AI alignment won't automatically solve as models grow

Microsoft CEO Satya Nadella said in a recent episode of the 20VC podcast that he has changed his view on AI alignment. He previously believed that as AI models become more capable, ensuring their behavior aligns with human values and goals would naturally become easier. He now says he no longer believes the two will develop in sync.

Nadella said the capability gains from larger models met expectations, but the alignment problem remains unsolved. He described alignment as a more intractable and different kind of problem. Earlier, on October 10, 2026, he posted on social media warning about AI risks and called for stronger external safety controls. He suggested companies should treat AI models as internal security threats, assume models could be compromised in their safeguards, and emphasized the need for reliable system design, human oversight, and operational procedures, as well as industry standards where existing ones are insufficient.

Nadella is not the only tech executive with such concerns. OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei have also warned about potential AI risks. In July 2026, more than 1,000 AI practitioners signed an open letter warning that AI capabilities may outpace human understanding or control.