Microsoft CEO Satya Nadella has proposed that companies treat autonomous AI agents as potential insider threats. He argues that no single model should simultaneously execute actions and verify them, as this prevents effective oversight.
During remarks on October 10, 2026, Nadella emphasized the need to separate intelligence from authority. He suggested implementing practical controls, including independent verification, transparency in reasoning, and external audits.
In July 2026, Nadella introduced the concept of a reverse information paradox, noting that AI interactions could lead to the leakage of proprietary corporate data.
Around September 14, 2026, Microsoft released a draft Code of Conduct for its MAI models, which prioritizes human oversight and prohibits deceptive AI behavior. In a mid-September memo, Nadella warned that inadequate safety measures could jeopardize a company's operational legitimacy.
