Microsoft CEO Satya Nadella has called for stronger safety measures for advanced artificial intelligence (AI) systems, warning companies against blindly trusting AI models that can access sensitive data and perform tasks independently. 

In a post published on October 10, 2026, Nadella proposed treating AI models as potentially compromised from the beginning and ensuring that businesses can monitor, restrict and stop their actions when necessary.

His recommendations come amid growing concerns about AI agents that can interact with websites, software and databases with limited human intervention. As companies deploy these systems for research, coding and business operations, questions about security, accountability and control have become increasingly important.

1. Treat AI Models as Potential Security Risks

Nadella argued that businesses should not assume an AI system is trustworthy simply because it was developed by a reputable company. Instead, organisations should treat advanced models as potential insider risks, particularly when they have access to confidential information or critical systems.

2. Introduce an Emergency Brake

One of Nadella’s key recommendations is to ensure that an authorised human can pause or shut down an AI model while it is performing a task. Such a mechanism would allow businesses to intervene if a system behaves unexpectedly, accesses restricted information or carries out an unsafe action.

3. Strengthen Monitoring and Independent Audits

Nadella also called for stronger monitoring, independent audits and reliable records of AI activity. These measures could help organisations understand how a model reached a decision, what actions it performed and whether it followed established rules.

4. Disclose AI Incidents Promptly

Another important part of the proposed framework is timely disclosure of significant AI failures and security incidents. Sharing reliable information about what went wrong could help businesses and researchers identify recurring risks and develop stronger safeguards.

Nadella’s recommendations highlight a shift in how organisations may need to approach AI adoption. As businesses give AI agents greater access to operational systems, security controls, human supervision and clear accountability will become increasingly important.

The proposals are recommendations rather than a universally adopted regulatory framework. Their broader impact will depend on how AI developers, businesses and policymakers translate these principles into practical safeguards.

Discover the next big investment! Tradz by EquityPandits’ IPO screener helps you identify promising initial public offerings. Download Tradz by EquityPandit and get ahead of the curve! Sign Up Now & Find Your Next IPO Gem!