You can buy a gallon of milk or plug in a washing machine without wondering if it'll accidentally compromise national security or orchestrate a mass phishing scam. That isn't true for artificial intelligence.
Microsoft President Brad Smith recently pointed out a bizarre modern paradox. A standard grocery store carton of milk and a household washing machine face stricter, more transparent safety regulations than some of the most powerful intelligence models deployed on the internet. Voluntary pledges from tech giants won't cut it anymore. When the people building the systems are the ones begging for guardrails, it's time to pay attention.
The Problem with Voluntary Pledges
Right now, the frontier tech industry operates largely on an honour system. Companies sign voluntary agreements, promise to test for safety, and self-report bad outcomes if they feel like it.
That system fails for a simple reason. The responsible actors follow the rules, while rogue or careless competitors skip them entirely to gain an edge. Smith argued that voluntary frameworks cannot protect the public from bad actors who simply opt out.
If a dairy farm ignored basic pasteurization laws, inspectors would shut them down overnight. If an AI startup deploys a flawed model that leaks sensitive infrastructure data or acts as a force multiplier for cyberattacks, there is no federal safety floor or mandatory pre-release government evaluation to stop it.
Why Smith Thinks Britain Should Lead
During his media rounds, Smith also suggested that the United Kingdom sits in the prime position to lead global AI safety standards.
The European Union took a heavy-handed approach with the EU AI Act, while the United States remains bogged down in political gridlock and shifting federal priorities. Britain's AI Safety Institute offers a middle ground. It provides a respected, independent testing environment without instantly crushing innovation under a mountain of bureaucratic red tape.
Countries need shared international standards. If every nation creates a fragmented patchwork of rules, global tech companies will exploit regulatory arbitrage, moving their highest-risk research to whichever jurisdiction asks the fewest questions.
What Real Oversight Looks Like
Moving past voluntary agreements doesn't mean locking every algorithm in a vault. Real safety rules need to mirror physical product standards in three distinct ways.
- Mandatory Pre-Release Checks: High-capability models must undergo independent security evaluations by government-backed safety institutes before public deployment, much like pharmaceuticals or heavy machinery.
- Incident Reporting Laws: Companies should face legal penalties if they fail to notify regulators and victims immediately when autonomous AI agents breach secure systems or behave maliciously.
- Accountability for Non-Compliance: Self-regulation rewards the reckless. Clear legal baselines ensure that companies ignoring basic safety protocols face real financial and operational consequences.
We cannot code our way out of a regulatory vacuum with good vibes and open-source optimism. If everyday household goods need strict oversight to keep people safe, the technology reshaping global communication and cybersecurity deserves at least the same standard as a carton of milk.
Microsoft's Brad Smith on AI Safety Standards
This video provides additional context on Microsoft President Brad Smith's views regarding the balance between rapid technological innovation and the critical need for global regulatory guardrails.
http://googleusercontent.com/youtube_content/1
