UK parliamentary committee recommends legislation allowing regulators to pause dangerous AI systems
Des députés britanniques proposent de suspendre les IA jugées dangereuses | Brief IA
A UK parliamentary committee recommended pushing legislation in the next parliamentary session to give regulators the power to halt the development or release of dangerous AI; the recommendation is non-binding and the government has not committed to adopting it. The committee said the UK AISI can currently only rely on companies voluntarily accepting evaluations and has no power to veto releases, so it proposed that an independent body be responsible for enforcement. For high-risk systems, developers would have to submit evidence of safety assurances and accept review by independent researchers, and would need to be re-evaluated when model capabilities or safeguards change significantly. The report also called for stronger legal protections for internal whistleblowers, mandatory reporting of incidents and near-misses, and listed ten shortcomings in the national system. The committee estimated that AI could bring up to £400 billion in incremental value to the UK economy over the next decade, and hopes to use the UK's 2027 G20 presidency to promote international standards.