Microsoft Chief Calls for Tougher AI Rules, Citing Milk and Washing Machines
Microsoft president Brad Smith says AI faces lighter oversight than everyday products and proposes new safety rules, including liability for victims of rogue agents.
Microsoft president Brad Smith has said that artificial intelligence is less regulated than milk and washing machines, according to The Guardian. He made the comparison as the technology industry voiced growing concern about the safety of AI systems. Smith called for more safety rules covering the technology, including a legal obligation toward a person who is harmed if their system is breached by rogue AI agents, The Guardian reports.
BBC News reports that Smith called for global rules on AI and said the United Kingdom is in the "best position" to lead that effort. The BBC's account frames his remarks as a push for international coordination rather than only domestic measures.
Both outlets describe the remarks as a warning about safety, and both place them in the context of a broader debate within the tech sector about how AI should be governed. Neither report, as summarized here, provides detailed proposals for how such obligations would be written or enforced.
The coverage focuses on Smith's statements rather than on any specific incident. The summaries available describe the concept of a victim of a rogue AI agent in general terms, without identifying a particular case, company or damages.
The reporting places Smith's comments within a debate over whether current rules keep pace with the speed at which AI tools are being deployed. Readers should treat the comparison to milk and washing machines as a rhetorical point made by one executive, not as a measured assessment of regulatory coverage.
Background
Regulation of consumer goods in many countries rests on long-established frameworks. Food products, household appliances and electrical equipment typically must meet safety standards, be tested or certified, and carry labeling or recall procedures. Those systems developed over decades, often after accidents exposed gaps, and they give regulators clear points of intervention when something goes wrong.
Software has historically been treated differently. Many digital products have been governed by contract terms, consumer protection law and sector-specific rules, rather than by a single product-safety regime. Liability for software failures has been difficult to pin down, partly because code is often licensed rather than sold, and because harms can be spread across many users, vendors and platforms.
Artificial intelligence adds further complications. Modern AI systems can be updated frequently, can behave unpredictably in new situations, and are increasingly deployed as autonomous agents that take actions on a user's behalf. Questions about who is responsible when such an agent causes harm, and what counts as a breach, do not have settled answers in most legal systems.
Governments have begun responding. Several jurisdictions have proposed or adopted rules that classify AI uses by risk, require transparency about training or outputs, or set obligations for developers of powerful models. These approaches differ in scope and enforcement, and international coordination remains uneven. Proposals for global standards have appeared in multilateral forums, though binding agreements are rare.
Industry leaders have often called for rules while also shaping what those rules look like. This is not unusual in technology policy, where companies have deep technical knowledge that lawmakers may lack. It does, however, raise questions about how to weigh industry input against public interests and about which voices shape the final text.
Why it matters
A call for a legal obligation to people harmed by AI agents goes to the heart of accountability. If a system acts on someone's behalf and causes damage, the question of who answers for it affects everyone who uses these tools, from individuals to small businesses. Clear liability rules could encourage careful design, while unclear ones may leave people with little recourse.
The comparison with consumer products also matters because it frames AI as something that should meet baseline safety expectations, much as a toaster or a dairy product does. Whether or not the analogy holds, it signals that a major technology company sees the current regulatory position as insufficient, which may influence debate in legislatures and standards bodies.
The mention of the United Kingdom as well placed to lead suggests that national positioning is part of the discussion. Rules set in one major market can shape products sold worldwide, so decisions about where standards emerge may affect which companies can comply most easily and which users are protected.
There is also a question of timing. Calls for rules from people inside the industry can be read as sincere concern, as an effort to set terms before competitors or regulators do, or as both. Readers and policymakers will need to judge the proposals on their details rather than on the messenger alone.
Finally, the story highlights a gap between how quickly AI tools reach ordinary people and how slowly the legal architecture around them develops. Closing that gap will require specific choices about definitions, responsibility and enforcement that are not yet visible in these reports.
What to watch
Watch for concrete proposals. A general call for a legal obligation to victims is a starting point, but the details will determine whether it changes anything: who counts as a victim, what counts as a breach, and who is liable.
Watch for responses from governments, particularly in the United Kingdom, and whether international bodies take up the call for global rules. Agreements across borders tend to move slowly and depend on broad political will.
Watch for reactions from other major AI developers and from consumer groups, which will show whether the industry speaks with one voice on safety rules.
Finally, watch whether any specific incidents involving rogue AI agents are reported, since concrete cases often shape the direction of regulation more than general warnings do.
Viewpoints
Commentary from our columns: three angles on this story.
The Skeptic open questions and weak points
It is easy to say AI needs tougher rules, and harder to say what those rules should require. The comparison to milk and washing machines is vivid, but it skips over how different these products are. A carton of milk does not rewrite its own instructions or take actions across a dozen services at once. Before accepting the analogy, readers should ask what exact standard would apply, who would test it, and how a regulator would measure whether a system is safe enough.
The idea of a legal obligation to a victim of a rogue agent also raises open questions. What counts as a breach? Who is liable when a system is compromised by an outside attacker rather than failing on its own? Those answers matter more than the headline. A call for accountability is welcome, but it becomes meaningful only when the terms are written down and tested against real cases. Until then, the proposal is a direction, not a plan, and the public should keep asking for the details.
The Ledger the business angle
When a leading software company calls for liability rules, it is worth asking how those rules would land on the balance sheet. Liability obligations can raise costs for developers, insurers and the businesses that deploy AI agents. Large firms may be able to absorb compliance expenses and legal risk more easily than startups, which could make tougher rules a competitive advantage for incumbents. That does not mean the idea is wrong, but it does mean the economics deserve scrutiny.
Companies also have a clear interest in shaping the rules that will govern them. Calls for standards can build trust with customers and regulators, and they can set terms that suit a firm's existing products. Readers should look at who would pay for safety, how insurance markets might price agent risk, and whether obligations would fall on the party best placed to prevent harm. The most useful debate will follow the money as closely as it follows the principles.
The Rulebook privacy, safety and regulation
Calls for AI safety rules raise practical questions about privacy, consent and accountability. If an AI agent acts on a person's behalf and that system is breached, the individual may have limited visibility into what happened or who was responsible. A legal obligation to victims could help people seek redress, but only if the law also requires clear records of agent actions, reasonable security standards and timely notice when something goes wrong.
Regulation should also avoid creating loopholes that let companies disclaim responsibility through complex terms of service. Any framework will need to define roles for developers, deployers and the businesses that connect agents to personal data. Readers should watch whether proposals address data protection, cross-border enforcement and the rights of people who never agreed to use these tools in the first place.