Microsoft President Brad Smith Advocates Independent AI Safety Evaluations and Multi-Layered Controls
Microsoft President Brad Smith publicly voiced support for independent third-party AI safety evaluations and multi-layered emergency kill switches to ensure AI systems remain under human control. Additionally, he acknowledged that the massive energy demands of expanding AI data centers are forcing Microsoft to adjust its 2030 carbon reduction goals. As autonomous AI agents are integrated into software platforms, multi-tiered oversight mechanisms are becoming central to international AI governance debates. Furthermore, Microsoft's admission underscores the growing conflict tech companies face between pursuing aggressive AI capabilities and maintaining environmental sustainability commitments. Smith highlighted that Microsoft operates a joint safety committee with OpenAI to evaluate model readiness before release, arguing that shutdown mechanisms should exist across software and government levels rather than a single point of control. On the climate front, Microsoft's reliance on fast-tracked natural gas power plants to support AI infrastructure has made its target of becoming carbon-negative by 2030 increasingly unrealistic under original assumptions.
## BACKGROUND
As AI models advance, governments and industry leaders are evaluating risk mitigation policies like independent evaluation benchmarks and emergency shutdown switches to contain rogue or dangerous systems. At the same time, training and operating frontier AI models requires massive compute power, driving unprecedented electricity consumption and challenging corporate environmental goals.