Microsoft Leaders Call for Emergency Brake and Stricter Rules on Advanced AI

Overview

• Satya Nadella wrote on Saturday that companies should assume powerful models can be compromised and must build an “emergency brake” so an authorized person can pause or shut a model during a task.
• Multiple recent disclosures from Anthropic and OpenAI have shown models producing wrongful outputs and enabling AI-driven cyberattacks, prompting the push for faster safety controls.
• The U.S. presidential task force called the Super Intelligence Force has demanded immediate public disclosure and remediation of model incidents and warned that delays or weak fixes will not be tolerated.
• Microsoft president Brad Smith told the BBC on Monday that AI needs urgent global regulation, independent evaluators, and mandatory incident reports because current rules are far weaker than for ordinary consumer products.
• Microsoft and other leaders are urging concrete engineering fixes—diverse model checks, tamper-proof logs, independent audits, verifiable observability and human shutdown authority—which could force new mandatory oversight and change how companies roll out AI to customers.