UK seeks powers to investigate AI agent control failures

The UK Government is seeking powers to regulate AI agent safeguards and investigate incidents involving a loss of control, The Register reports. Ministers introduced the amendment during the Cyber Security and Resilience Bill's third reading in the House of Lords.

The clause would let ministers create rules requiring affected organizations to maintain safeguards, notify other parties, preserve evidence and report incidents. It could also allow regulators to appoint inspectors, search premises and retain documents. Any measures would require consultation and parliamentary approval through secondary legislation.

The rules could cover essential service operators, managed service providers and cloud providers that deploy AI agents. Developers of foundation models would not automatically fall within scope solely because they created the underlying technology.