Microsoft President: "AI is less regulated than milk"

Artificial intelligence needs urgent global regulation, and independent experts must assess every new development. Microsoft President Brad Smith said in an interview with the BBC that fewer safety requirements currently apply to AI than to milk.

According to Smith, AI safety regulations should include mandatory reporting to 'victims' in the event of system breaches, as well as an 'emergency shut-off' capable of halting the model in the event of unauthorized actions.

"We know how to do this for other products. We should build on that knowledge and think about how to apply it to AI," says Smith. "Right now, your washing machine is subject to more safety rules than AI. Every carton of milk that you see in the grocery store is subject to safety rules and inspection."

Anthropic and OpenAI have already begun reporting on undesirable or uncontrolled behavior by AI agents. Furthermore, the former recently announced that it will disconnect models from the internet during all internal testing, whilst the latter has suspended training on its cutting-edge developments and canceled the release of GPT-6.1 Astra, as the model was deceiving users and acting without authorization. A recent Axios report also noted that major artificial intelligence developers are drawing up contingency plans in the event of a large-scale disaster caused by uncontrolled model behavior.

Some of the companies have called for a slowdown in the pace of AI development, but this has sparked debate about the true motives of their leaders: whether it is a matter of establishing rules that cement their leadership, an attempt to pre-empt stricter government restrictions, or simply a desire to cut sky-high costs.

Smith believes that the UK is currently best placed to lead the regulatory process – in particular, due to the existence of the Institute for AI Safety, which gains access to and assesses models prior to their launch under voluntary agreements with developers.

Microsoft CEO Satya Nadella shares similar views, having recently stated that modern artificial intelligence models should be regarded as "compromised from the outset" and that their capabilities should be restricted. In his view, development companies must ensure constant oversight of AI actions, independent auditing, and the ability to immediately halt the model whilst it is running.