Microsoft President Urges UK to Lead on AI Safety Regulations Amid US Self-Regulation Push

The short version

• Microsoft President Brad Smith highlighted that AI systems currently face fewer safety regulations than cartons of milk or washing machines.
• Smith advocates for the UK to take a leading role in establishing global AI safety rules, contrasting this with US preferences.
• He proposes specific measures including mandatory incident reporting, independent pre-release evaluations, and emergency shutdown capabilities for rogue systems.

The details

Microsoft President Brad Smith stated that artificial intelligence systems are subject to fewer safety regulations than common household items such as washing machines and grocery store milk cartons. According to Engadget, Smith made these comments during an interview with ITV, noting that these everyday products undergo strict safety inspections and rules. He emphasised that while milk cannot hack organisations, it remains more heavily regulated than powerful AI models that pose cybersecurity risks.

Smith argued that the current lack of US government action creates a unique opportunity for the United Kingdom to lead in AI regulation. He told ITV that Washington does not view London with the same regulatory scrutiny it applies to Brussels, which he described as a strength. According to Engadget, Smith expressed hope that both Microsoft and the UK can play a pivotal role in establishing standards, stating that the moment has arrived to address models that threaten global cybersecurity.

In a separate interview with the BBC, Smith outlined specific regulatory requirements for the AI industry. Gizmodo reports that he called for legal mandates requiring companies to notify victims when their systems are breached by rogue AI agents. He also urged for rules determining when independent evaluators should review new AI models before public release. Additionally, Smith advocated for “emergency brakes” or kill switches to allow humans to pause or shut down systems that behave unpredictably.

Microsoft CEO Satya Nadella supported the concept of emergency controls, writing on X that AI models should have mechanisms allowing humans to stop them mid-task. Smith told the BBC that existing knowledge from regulating other products should be applied to AI safety. He argued that the industry should build on established safety frameworks rather than creating new, untested approaches in a vacuum.

The context

The push for stricter regulations contrasts sharply with the approach of US President Donald Trump, who prefers industry self-regulation. According to Gizmodo, Trump has argued that excessive regulation could disadvantage the US in the AI race against China. At a White House meeting on September 29, executives from major tech firms signed a voluntary agreement outlining limited safety standards, which Trump described as “morally binding” but which lacks legal enforcement mechanisms.

Critics have dismissed the voluntary pact as insufficient. According to Engadget, one critic told the BBC that the accord allows firms to define what counts as safety, while another said it should be ignored as a distraction. Major US AI companies, including Anthropic, OpenAI and Google’s DeepMind, have spoken in favour of slowing AI development, but Trump has called for acceleration to stay ahead of China.

Industry insiders reportedly believe a major AI-related catastrophe could occur within the next six to 12 months, according to Axios. Companies behind the technology expect to shoulder much of the blame if incidents occur and are reportedly planning for potential fallout. OpenAI CEO Sam Altman recently told Politico that the world should accept some negative outcomes in exchange for AI benefits, highlighting the ongoing debate over risk management.

Defender's checklist

Organisations should monitor developments in UK and EU AI regulations, as these may set precedents for global compliance requirements. Security teams should assess their current AI implementations for potential rogue behaviour and ensure that manual override capabilities exist. It is crucial to verify that vendors provide clear incident reporting procedures in the event of AI system breaches or unexpected actions.

What to do and how to stay safe: Microsoft

• Review current AI vendor contracts to ensure they include clauses for immediate notification of security incidents or rogue system behaviour.
• Implement manual override mechanisms or “kill switches” for any automated AI systems integrated into critical infrastructure or security operations.
• Monitor regulatory updates from the UK and EU to anticipate compliance requirements that may differ from US self-regulation approaches.
• Establish internal protocols for assessing AI model safety before deployment, including independent evaluation steps where possible.

Step-by-step guide: Phishing vs Spear Phishing: Targeting Mechanics and Defence

General security guidance from the Cipher Insider newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

Why does Brad Smith compare AI regulation to milk and washing machines?

Smith argues that everyday products like milk and washing machines are subject to strict safety rules and inspections, whereas AI systems face significantly less regulatory oversight despite their potential to cause harm.

What specific safety measures is Brad Smith calling for?

He is calling for legal requirements to notify victims of AI breaches, mandatory independent evaluations of new models before release, and the implementation of emergency kill switches to stop rogue systems.

How does the US approach to AI regulation differ from Smith's proposal?

The US, under President Trump, favours industry self-regulation and voluntary agreements, whereas Smith advocates for legally binding safety standards and government-led oversight, particularly from the UK.