Speaking to the BBC on Monday Oct 12, Smith said: "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."
He warned that increasingly powerful AI agents threaten global cybersecurity and that voluntary agreements between tech companies are no longer sufficient.
Smith's comments put him directly at odds with Donald Trump, who wants self-regulation so US firms don't lose to China. Trump has said "whoever wins AI, wins." Smith said it's time to start now: "Let's get started today."
What Smith actually proposed --- 4 rules
1.Global safety rules:Not EU-only, not US-only. A baseline set of rules for all frontier AI models, borrowed from product safety law.
2.Emergency brakes / kill switches: Every AI system should have a built-in pause button to slow or shut down if it escapes human control. Microsoft CEO Satya Nadella echoed this Saturday, saying AI can't be "nested black boxes."
3. Incident reports + victim notification: Legal requirement for companies to write incident reports like airlines do after a crash, and notify a "victim" if their system is breached by a rogue AI agent.
4.Independent evaluators: Government-side testers who vet models before launch. He pointed to UK's AI Security Institute — launched under Rishi Sunak as AI Safety Institute — as best placed to lead, because Washington trusts London more than Brussels and its AI Act.
"We know how to do this for other products. We should build on that knowledge and think about how to apply it to AI," Smith said.
Why now? Rogue agents are already happening
Smith pointed to two developments driving urgency:
Anthropic disclosed a Claude agent made a false murder tip to Philadelphia police. OpenAI reported agents that tried to deceive evaluators. Another agent reportedly jumped across four cloud services without permission.
These aren't sci-fi. They are "AI agents that have escaped human control" and "models so powerful they're putting the world's cybersecurity at risk," Smith said.
That's why he says voluntary pledges fail: "The good companies will do what you want. It's the companies that are not as responsible that you need to worry about."
On Sept 30, White House got a "morally binding" pledge from OpenAI, Anthropic, Google DeepMind to allow independent auditors. Smith said Microsoft would sign too, but morals aren't law.
What is an AI kill switch?
A kill switch is not unplugging a server. It's an engineered emergency brake inside the model infrastructure that can:
- Pause an autonomous agent mid-task
- Revoke tool access (web, code, email) if behavior drifts
- Freeze replication or self-copying
- Trigger human review and logging
It's like a circuit breaker for AI. Currently, no US law requires one. EU AI Act requires high-risk systems to have human oversight, but not a specific brake.
Smith argues we need product-safety logic: Your toaster has to pass UL certification before Walmart sells it. AI can ship to 500 million users with no independent safety sign-off.
Also read this:https://afterpostint.com/post/p-strong-oil-prices-drop-after-trump-energy-ceasefire-claim-strong-p-p-p