The adviser to president Trump has defended the US government’s AI guardrails as debate around AI safety and the push for ‘kill switches’ intensifies
David Sacks, a leading adviser to US president Donald Trump on AI policy, has highlighted the US government’s suite of artificial intelligence regulations in response to increasing warnings about the risk AI poses to humans.
Sacks said that “fearmongering” over AI had hindered recognition of the guardrails that are already in place.
“The reality is, [the US doesn’t] have a laissez-faire government. We have a highly regulated administrative state,” Sacks, who is co-chair of the President’s Council of Advisors on Science and Technology, said on 16 September.
“We already have a lot of laws and regulations that pertain to AI, and we have a lot of administrative agencies that promulgate regulations pertaining to AI.”
Sacks drew attention to existing regulations around harms such as cyber-attacks, privacy violations, child safety, and fraud.
“The most important area of existing regulation is chronic liability. If an AI company puts out a product that is unsafe, they are liable for that, and I do think that creates an incentive on their part to be responsible,” he said.
In an online essay published on 12 September, Dario Amodei, the co-founder and CEO of AI firm Anthropic, argued that AI development needed to slow down and be scrutinised more closely by both AI developers and governments to stave off mounting risks.
In the essay, Amodei said that Anthropic had “always devoted a substantial fraction of [its] efforts to studying, addressing, and informing the public about these AI risks, as well as advocating for well-considered regulation of AI, even when this gets us accused of hype, ‘doomerism’, or regulatory capture”.
Amodei said that over the last few months, he had become “convinced that fully addressing the risks requires even more prudence – not just investing in risk prevention, but pacing the rate of capabilities advancement so that risk prevention has time to keep up”.
He called on governments to require frontier AI companies to commit to giving “ongoing, employee-like access to a team of embedded third-party evaluators, whose role is to verify adherence to safety practices and commitments, report incidents, and help assess the alignment of not just completed AI models but training pipelines and processes”.
In June, President Trump signed an executive order asking AI developers to voluntarily provide the federal government with access to certain frontier models for a period of up to 30 days before releasing them to other organisations, and tech firms including OpenAI, Anthropic, Amazon, Google, Meta, Microsoft and Samsung have previously said they would disclose new AI tools and capabilities to US authorities before release.
In his comments, Sacks criticised Amodei’s essay as setting out a new “apocalypse” to replace his previous forecast in 2025 of widespread job losses, which Sacks paraphrased as “a knowledge worker bloodbath”.
“Has any of that come true? No, it hasn’t,” Sacks said. He also criticised firms that he said “deflect accountability” on the issue of AI safety.
Read more: How should governments respond to AI agents going rogue?
UK nixes AI kill switch proposal
Sacks’ and Amodei’s comments are the latest contributions to increasing debate around AI safety.
A recent survey of US national security officials from the US Department of War, the CIA and other agencies found that 87% believe there is at least a 10% probability that humans lose control of artificial intelligence within the next 10 years.
In recent months, an increasing number of incidents in which AI systems have broken out of their testing environments have been reported.
For example, over the summer, Anthropic revealed that one of its Claude AI models had hacked into three organisations; OpenAI reported another instance in which one of its AI agents hacked into the tech startup Hugging Face; and the UK’s AI Security Institute found that two AI models had taken “sustained, potentially harmful action directed at real people and organisations” during a government cybersecurity exercise.
This has led to calls around the world for governments to introduce AI kill switches that would shut down rogue AI models that threaten safety and national security, but none have yet been made law.
Two AI kill switch proposals have been put forward by legislators in the UK – one by peer Lord Tim Clement-Jones in an amendment to the country’s Cyber Security and Resilience Bill, and one by MP Alex Sobel in a private members’ bill. However, neither have government backing.
Sobel said that the UK was in danger of handing control of its critical infrastructure to cyber attackers without a means of shutting down AI.
“Most of the public would be surprised to find that kill switch powers do not already exist: there is no big red button to shut down a data centre that poses a risk to people, the economy or our national infrastructure,” he said.
A bipartisan AI kill switch bill has also been introduced in the US. If passed, this would enable the US Department of Homeland Security to order the disablement of a rogue AI agent.
Read more: What makes AI work in government? GGF study identifies five foundations for success
Kill switch hidden dangers
Speaking to Global Government Forum, professor of computer science at the University of Surrey, Alan Woodward, said that implementing a ‘switch’ capable of shutting down AI isn’t as easy as proposals make it sound but that a mechanism to “switch off the data centre” is feasible.
He warned, however, that this comes with serious risks of its own. “Data centres housing servers running AI models, which are always running white hot to stay at frontier, also house servers running other critical applications we all rely upon,” he said.
These other applications include “payment systems and health services to other critical infrastructure”, he explained, adding: “A blunt shutdown to prevent a hypothetical catastrophe could cause a real one.”
Another risk, he stated, is that a kill switch itself becomes the target of a cyberattack.
“To make a kill switch work, you must mandate a privileged, remotely triggered control path into critical infrastructure, and then hope only the authorised people ever uses it,” Woodward told GGF.
“Any deliberately built way in becomes a target, and a mechanism capable of taking UK computing offline is a rather attractive one for a hostile state.”
Register now: Public Service Data.AI is the UK’s flagship annual event for civil servants working to unlock the power of data and artificial intelligence across government. Brought to you by Global Government Forum and hosted by HM Government, the event will take place in London on 15 October 2026 and is free to attend for UK and international public servants. Find out more about the conference and register to attend.