The UK’s Joint Committee on Human Rights is urging a pre-deployment approval system for AI tools judged high-risk, arguing that post-deployment fixes are often inadequate. Regulators told the committee there is no established framework to pre-assess such models, and lawmakers warn powerful AI is being deployed faster than rules can be written. Industry figures including Anthropic’s Dario Amodei—and supporters such as Sam Altman and Elon Musk—have called for a slower pace to allow better oversight. The proposal would target only high-risk systems and could require developers to identify risks, test against safety standards, and provide evidence before deployment.
UK Lawmakers Push For Pre-Deployment Approval Of High-Risk AI Systems
Front view of the British Parliament building and the Big Ben in the UK | Image Credit: Wilfredor, Wikimedia Commons
UK lawmakers are increasingly calling for a system that would allow regulators to scrutinise AI systems judged to be high-risk before those systems are deployed. They say fixing problems after an AI system is live is becoming harder as models grow more capable and more widely used.
What Is Being Proposed
The proposal would require a form of pre-deployment assessment for certain high-risk AI models, shifting some responsibility from post-deployment oversight by existing sectoral regulators to an earlier stage in the product lifecycle. So far, the UK has relied mainly on established regulators rather than a centralized AI-specific approval regime.
Why Regulators Are Concerned
The Joint Committee on Human Rights has been examining whether current safeguards adequately protect privacy, equality and other rights in the face of AI risks. During its inquiry, regulators told the committee they had not identified a ready-made framework for pre-assessing high-risk AI models.
Committee members pointed to the example of live facial recognition, which first emerged in the UK in 2016 and took years for authorities to investigate and regulate—an experience many see as a warning about the pace of AI oversight.
Speed Versus Oversight
Lawmakers warn that increasingly powerful AI systems are being introduced faster than regulators can assess them or develop appropriate rules. The committee is debating whether current technical testing and legal protections are sufficient when AI can affect large numbers of people quickly and at scale.
Voices Calling For Caution
Dario Amodei, CEO of Anthropic, wrote on September 12:
"We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain."
Amodei warned that recent instances of recursive self-improvement could outpace researchers' ability to understand and control systems if left unchecked. Notably, leaders at rival AI firms—including OpenAI’s Sam Altman and xAI’s Elon Musk—have publicly supported calls to temper the pace of development, an unusual convergence given their competitive positions.
Current UK Efforts And International Context
The UK’s AI Security Institute has worked with experts to design pre-deployment testing and evaluations, including exercises assessing systems from companies such as Anthropic and OpenAI. However, the UK government has not adopted a formal pre-approval model and generally favors regulation at the point of use.
By contrast, the European Union’s AI Act already requires certain high-risk systems to undergo third-party conformity assessments before they are placed on the market or put into service.
Potential Features Of A Pre-Deployment Regime
- Targeted Scope: Only systems judged capable of causing serious harm would be subject to pre-deployment scrutiny, not every chatbot or feature.
- Risk Identification: Developers would need to identify and document potential harms.
- Testing And Evidence: Systems would be tested against defined safety requirements and developers would provide evidence showing compliance.
- Regulatory Enforcement: Regulators would gain clearer legal grounds to block or require fixes for systems that fail to meet requirements.
Outlook
It remains uncertain whether the proposed UK pre-deployment regime will become law. But the debate has intensified over whether some high-risk AI systems should be required to demonstrate they are safe before deployment—balancing innovation, public safety and the practical capacity of regulators.
Help us improve.
