Microsoft Pushes for Third-Party AI Safety Evaluators
AI-generated image · Bay Street Wire
President Brad Smith argues that AI safety requires multiple layers of oversight beyond the primary model developers.
Microsoft Corp. president Brad Smith has expressed support for the use of independent evaluators to oversee artificial intelligence safety, according to reporting from Bloomberg News via the Financial Post.
Speaking in an interview with Bloomberg Television during the United Nations General Assembly in New York, Smith argued that AI safety cannot progress if it is managed by only one or two entities. He specifically referenced leading U.S. AI labs Anthropic PBC and OpenAI. Smith noted that while the companies creating the models are critical, firms like Microsoft—which produce the software layers that monitor and control AI agents—are equally important to the process.
Smith detailed that Microsoft has already collaborated with OpenAI on safety discussions through a joint safety board tasked with determining when models are safe for release. He further advocated for a multi-layered approach to safety, including the implementation of an "off-switch" to maintain human control over advanced systems. Smith suggested that such a switch should not be centralized, noting that government entities could potentially hold one of several available controls.
In the same interview, Smith addressed other corporate initiatives, confirming Microsoft's plan to use 200 megawatts of computing capacity at a data center in the United Arab Emirates. This follows a US$1.5 billion investment in Emirati tech firm G42 and an announcement that Microsoft will spend an additional US$2 billion in the Middle East. Smith also acknowledged that these expansions would require adjustments to Microsoft's 2030 climate goals regarding carbon emissions.