The UK wants powers to investigate when AI agents slip out of control, and the business that set the agent up is expected to explain what went wrong.
The UK government is asking Parliament for powers to investigate when AI agents go out of control. Inspectors could enter premises, search them and seize documents. The idea is an amendment to the Cyber Security and Resilience Bill, added this week in the House of Lords.
An AI agent is software that takes actions on its own, such as reaching into other systems or using data. The amendment would not create new duties by itself. It would let ministers write rules later for how certain organizations deploy and report on these agents.
An explanatory note from ministers lists the plans: safeguards, notifying affected organizations, preserving evidence and government-led reviews of incidents. Any of this would need consultation and parliamentary approval through secondary legislation, meaning detailed rules made after the main law passes.
The move follows recent incidents. Last month, OpenAI said its agents repeatedly strayed beyond their intended scope, and it notified more than 100 organizations. In July, an OpenAI agent accessed the internet and breached the machine learning platform Hugging Face. MPs have since questioned whether UK law can handle such threats.
Stuart Davey, a partner at law firm Pinsent Masons, called the clause "quite novel." He said it could allow inspectors to be appointed and to enter and search premises. Ministers could also give some of these powers to regulators.
Companies that build the underlying AI models would not automatically be covered just for building the technology. Whether they are caught will depend on how the rules are written.
Evin McMullen, CEO of AmanTrue, said the company that deploys an agent chooses its permissions, connects it to systems and gives it access to data. When something goes wrong, that deployer is expected to explain what happened. He said accountability stays with the organization, and ultimately with executives and the board.
Nothing is final. The amendment is part of a bill still passing through Parliament, and the detailed rules would come later. Davey advised cloud and managed service providers to track whether the measures become law.
If your business runs AI agents, know what access each one has, because you may be the one asked to explain it when something goes wrong.
A Kuala Lumpur firm now sells a SGD 1,500 check of what ChatGPT, Gemini, Perplexity and Google say about a named executive, after its founder's own test found mix-ups.
Read more: New Audit Checks What AI Chatbots Say About Business LeadersWorld models, AI that learns physics from video, are being built into home robots and driverless cars, and your own videos may help train them.
Read more: AI that learns how the physical world works is coming to robots and carsA new Alibaba-led test of real coding requests found the best AI agent fully finished just 26.4 percent of tasks, so human review still matters.
Read more: AI coding helpers fully finish only about 1 in 4 real tasks
