Ministers weigh compulsory AI safety reporting for UK developers
Sifted reports that the UK government is considering making AI safety reporting mandatory. The shape of any scheme — who files, what they file and when — has not been set out.
The UK government is considering compulsory AI safety reporting, according to a report by Sifted.
The report ties the thinking to Kanishka Narayan, who is named in Sifted's coverage. Nothing has been announced as policy.
Safety reporting in the UK has so far run on voluntary commitments from large model developers. Making it compulsory would move it from a goodwill exercise to a filing obligation, and filing obligations have thresholds, deadlines and penalties attached.
Those details are the whole story for a startup, and none of them exist yet in public. No threshold has been published that would tell a founder whether a small team fine-tuning someone else's model is in scope or out of it.
Nor is there a published list of what a report would have to contain, or how often it would have to be filed.
For founders building on AI, the practical move is unglamorous: keep a record of what your models do, what you tested, and what you found. Firms that already document evaluations have a cheap compliance exercise ahead of them. Firms that do not will be reconstructing a year of work from memory.
Consultation is the usual next step for a measure of this kind. Until one opens, there is nothing to respond to.