Microsoft's Brad Smith Backs Independent AI Safety Review and Human Control Safeguards

During the UN General Assembly in New York, Microsoft President Brad Smith said in a Bloomberg interview on September 23 that he supports bringing in independent evaluators to assess AI safety. Smith said that if AI safety is handled only by one or two organizations, it will be hard to make progress. Model companies are crucial, he noted, but companies like Microsoft that provide the software layer, control AI agents and monitor agent behavior are equally important.

Smith said Microsoft and its close partner OpenAI have a safety committee with members from both companies. Microsoft takes part in AI safety discussions through the committee and helps judge when a model is safe enough to release. He also said Microsoft supports kill switches for advanced systems to keep them under human control. Safety mechanisms need to be deployed at multiple levels, he said, adding that if switches can be placed in three or four places, they should not be placed in only one.

Smith also acknowledged that Microsoft needs to adjust its climate goals. The company previously committed to removing more carbon from the atmosphere than it emits by 2030. That timeline appears increasingly difficult to meet, in part because Microsoft has used natural gas power plants that can be quickly brought online to supply AI data centers. Smith said climate remains one of the most important issues of this century, but the question is whether these goals can be achieved in the way envisioned six years ago, and some adjustments are necessary.