# UK government bets on evidence over mandates for AI risk > Source: > Published: 2026-09-14 11:42:44+00:00 Photo: Tima Miroshnichenko / Pexels # UK government bets on evidence over mandates for AI risk Andy Burnham's administration backs voluntary testing and transparency, rejecting calls for hard bans or a statutory kill switch The UK government has drawn a clear line in the sand on artificial intelligence regulation: it wants companies to be honest about risks, it wants policy rooted in science, and it does not want a kill switch. Prime Minister Andy Burnham’s spokesman confirmed that the government supports AI companies being open about the dangers their models pose, and that any future regulatory measures will be grounded in empirical evidence rather than precautionary bans or rigid mandates. ## The framework taking shape The Cabinet Office made the government’s position explicit: policy decisions on AI will be anchored in scientific evidence, and responsibility for safe development sits with the companies building these systems. Parliament has been pushing for harder measures, including proposals for a statutory mechanism to shut down rogue AI models and, more dramatically, a potential ban on superintelligent AI altogether. The government has declined to endorse either. The rejection of a so-called kill switch is particularly revealing. The government’s argument is that such a mechanism would constrain innovation without actually reducing risk. For context on those stakes: Anthropic researcher Evan Hubinger has raised the possibility of a greater than 10% chance of species-level risk from AI within a decade. That figure surfaced in parliamentary discussions as Burnham addressed the Commons on September 9, 2026. The government’s preferred instrument is the AI Security Institute, the rebranded successor to the AI Safety Institute that emerged from the 2023 Bletchley Park summit under then-Prime Minister Rishi Sunak. The AISI evaluates frontier models before they reach the public, operating under voluntary access agreements with developers. Voluntary is doing a lot of work in that sentence. Anthropic recently withheld its latest model from AISI testing, a decision that created friction in Westminster and underscored the central tension in the government’s approach: a framework that depends on company cooperation has limited leverage when a company decides not to cooperate. ## Innovation-friendly, but not consequence-free The UK’s broader regulatory posture is what policymakers sometimes call risk-proportionate, meaning the severity of intervention scales with the severity of the threat. In practice, this looks like targeted assessments, pre-release evaluations, and international coordination rather than sweeping legislation. The government has been explicit that it favors international cooperation as a complement to domestic policy. The groundwork for that came from Bletchley, where Sunak assembled a coalition of governments and major AI labs for the first major global AI safety summit. Burnham’s administration has inherited and continued that architecture, even as the institutional branding has shifted toward security rather than safety, reflecting the growing overlap between AI risk and national security concerns. For AI companies operating in the UK, the current environment is relatively permissive. Pre-release testing is expected but not legally required. There is no mandatory disclosure regime, no liability framework for AI-caused harms, and no hard cap on model capability. Parliamentary pressure is not going away. As of mid-September 2026, calls for stronger legislation have been building, and the government’s continued reliance on voluntary agreements gives opponents a clear attack surface. **Disclosure:** This article was edited by Editorial Team. For more information on how we create and review content, see our [Editorial Policy](https://cryptobriefing.com/editorial-policy/).