A parliamentary committee in the United Kingdom has issued a formal recommendation for the creation of a new AI regulatory body equipped with the authority to halt the development or deployment of high-risk artificial intelligence models. The report, titled Guardrails for Growth: Making AI Safe, argues that the current reliance on voluntary safety commitments from major technology companies is insufficient to protect the public from the unpredictable nature of large language models.
The Business, Innovation, Science and Trade Committee, which conducted the inquiry, heard testimony from representatives of major AI firms including OpenAI, Google DeepMind, Anthropic, and Meta. While these companies maintained that they do not release models they deem unsafe, the committee concluded that there is no standardized industry yardstick for safety and no legal requirement for independent verification before a product reaches the market.
The proposed AI Safety and Security Bill would establish a regulator capable of intervening during the development and testing phases of AI systems. This watchdog would have the power to block the deployment of models, halt high-risk training processes, or suspend existing services if safety risks cannot be adequately mitigated. The committee emphasized that because the most advanced models are trained on billions of parameters, developers often cannot fully predict how these systems will behave in every scenario.
Currently, the UK AI Safety Institute (AISI) conducts testing on powerful models, but its operations are limited by the voluntary nature of its cooperation with developers. The committee recommends that the AISI continue its technical evaluation work, but that a separate, empowered regulator be established to handle enforcement and legal oversight. Under the proposed framework, developers of high-risk systems would be legally required to provide evidence supporting their safety claims and grant independent researchers access to examine their models.
The report identifies ten specific deficiencies in the current British approach to AI governance. Among these are the lack of mandatory incident reporting, insufficient protections for whistleblowers, and limited access for independent auditors. The committee noted that former AI employees have reported facing commercial pressures, financial disincentives, and restrictive contracts that discourage them from raising concerns about safety practices.
To address these issues, the committee is calling for stronger legal safeguards for whistleblowers and a mandatory system for reporting AI-related incidents and near misses. The report suggests that by formalizing these requirements, the UK could foster greater trust in the technology, potentially encouraging smaller businesses to adopt AI tools. The committee estimates that the economic benefits of AI could increase the British economy by as much as £400 billion over the next decade.
The committee has urged the government to introduce the necessary legislation during the upcoming parliamentary session. Furthermore, the members suggest that the UK should leverage its leadership position during the 2027 G20 presidency to advocate for international standards regarding AI testing, incident reporting, and the release of high-risk models.
It remains unclear whether the government will adopt these recommendations. The committee’s findings are not binding, and ministers have not yet committed to drafting the proposed legislation. The debate centers on the balance between fostering economic growth through innovation and the necessity of state intervention to manage risks that, once released into the public domain, may be impossible to recall.





