Microsoft president urges stricter AI safety rules than milk

Brad Smith’s comments are notable because they come from the president of one of the firms most aggressively deploying frontier models, not an external critic. When Microsoft’s top lawyer says AI is less regulated than a carton of milk and explicitly calls for kill switches and liability to victims, he is effectively inviting lawmakers to hard-code constraints on how AGI-era systems are built and operated. This is less about today’s chatbots and more about pre-positioning for a world where autonomous agents control physical systems, financial flows and critical infrastructure.

The subtext is that Microsoft wants predictable, cross-jurisdictional rules that lock in the compliance capabilities it has already invested in. Smaller labs and open-weight communities will find it harder to implement audited kill switches, red-team programmes and legal compensation frameworks at the same level. That tilts the playing field toward incumbents who can absorb the compliance overhead. At the same time, Smith’s framing that AI can be both "tool and weapon" validates fears in Brussels, London and Washington that frontier labs cannot be trusted to self-regulate indefinitely.

For the AGI timeline, this raises the probability of hard regulatory guardrails on deployment and perhaps even training of very large models. Rather than slowing fundamental research, such guardrails are more likely to channel it toward verifiable safety properties, secure model hosting and controlled interfaces for high-risk use cases.