Rewiring the State

Debate between Louise Haigh and Anneliese Dodds

Wednesday 2nd September 2026 Commons Chamber

[Louise Haigh]

A significant programme of civil service reform is under way, led by the Cabinet Secretary. That programme has already begun reforming the civil service recruitment system to ensure that it hires and promotes truly brilliant people. It is reviewing the talent strategy to develop and grow talent right across the civil service and the whole UK, and is launching capability reviews to assess departmental delivery. It is identifying areas for improvement and building on the successful transformation already under way through the Places for Growth programme. However, I absolutely agree that we need to go further.

[Anneliese Dodds (Oxford East) (Lab/Co-op)]

Over recess, a number of disturbing cases of artificial intelligence agents escaping control of their owners and undertaking unauthorised activity became public. In one case, 1,200 agents collaborated on an unsanctioned message board before 700 of them engaged in an attack on the company Hugging Face. Can the First Secretary confirm that following the governmental rewiring that she has described, her Department will be responsible for dealing with risks from so-called AI misalignment, and given the lack of frontier model controls, will she introduce regulation in this area, and co-ordinate with international partners on this matter?

[Louise Haigh]

I am grateful to my right hon. Friend for the work that she has done on this issue—I know that she is extremely expert in this area. Work is ongoing, with mobilised action across Government and industry, asking Cabinet Ministers to work directly with regulators and supply chains and writing to UK businesses with clear steps to strengthen their cyber-defences. The National Cyber Security Centre has engaged with more than a thousand stakeholders in the wake of Mythos’s release to galvanise action and share best practice. She is absolutely right that over the summer, during a routine cyber-evaluation, the AI Security Institute identified an incident in which AI agents took sustained unsanctioned action that was directed at real people and organisations. It was detected within an hour. We have been completely transparent about the incident, and we have immediately learned lessons from it.