Andy Burnham needs to change government policy on AI

That means ditching the one that was shaped by Peter Mandelson, argues Frank Hansen. 

On 10th September, Elon Musk mocked claims by ex-Anthropic researcher Jacob Coxon that super-AI development (“Frontier Models”) posed an existential danger to humanity. Coxon had accused Anthropic and OpenAI of “racing straight to self-improving, superintelligence and gambling with our lives”, and called for a slowdown. Musk replied that this could be a conspiracy or a public relations stunt, linked to demands by Democrats for stricter AI regulation.  

It’s strange that Musk should say this as he has often argued that AI-driven human extinction is a statistical probability, famously estimating a 10% to 20% chance. Even stranger, perhaps, is that within a few days Musk flipped and with OpenAI’s Sam Altman, and Google DeepMind’s Demis Hassabis backed a call by Anthropic boss Dario Amodie to “slow the pace”, warning that an AI bot swarm could be capable of “taking over the entire internet” within a year. Amodie called on Washington to coordinate control of AI systems, including internationally with China.   

Meanwhile Donald Trump was sticking to the line that it was all a “sick conspiracy” and full steam ahead was the only way to win the AI race with China. He said that the only guardrail needed was himself since he was “strong and smart”.  JD Vance chipped in later with his ‘Frankenstein speech’, acknowledging that AI could be a problem, but saying companies should sort it out for themselves: “If you’re gonna create Frankenstein, don’t come to the government and say we need regulation.”  Essentially, these two want to continue the current policy of voluntary self-regulation and are opposed to the global regulation of AI.

However, in the face of calls for a slowdown from people as diverse as Bernie Sanders and Steve Bannon, plus rising public and local state opposition to hyperscale data centres across the US, these four Tech Bros see the need to respond in some way and issued their ‘Deceleration Pact. Of course, we know these oligarchs have ulterior motives and it’s difficult to believe anything they say. 

The joint statement is not exactly typical of their previous ‘cordial’ relations. In 2015, Altman and Musk founded OpenAI and Amodie soon joined. Musk left following a power struggle in 2018 and went on to unsuccessfully sue the company for $150 billion. He founded xAI and Grok in 2023. Amodei left in 2020 citing a profound breakdown in trust with Altman’s leadership and proceeded to set up Anthropic.  The three now lead rival companies that have rapidly evolved from experimental labs into multibillion dollar US AI firms which, along with META, Google and Apple, are engaged in a ruthless race to control a potentially lucrative world market. It certainly seems to be a real crisis point in the AI race.

The pact is mainly to do with PR ‘concessions’ and commercial interests of course. Rising demands for safety regulation pose a threat and they want to manage the debate to ensure it doesn’t affect their competitiveness or future profitability. They also want to calm the markets – OpenAI and Anthropic were on the brink of going public and issuing shares with hopes of raising up to $2 trillion.  Altman has confirmed that the issue will not now happen in 2026. He is a very smart PR operator and immediately disclosed six more examples of “unexpected or concerning” behaviour by OpenAI technology, and said the company was introducing a new framework for investigating and disclosing AI model ‘misalignment’, the term for AIs failing to adhere to human values and safety goals. 

At the same time, OpenAI’s partner Microsoft, while supporting the urgent need for safety controls, reverted to commercial rivalry by launching a strong attack on its main competitor Anthropic. Microsoft AI blamed Anthropic for developing AI “like it is human”, which could lead to the emergence of a new “silicon species” competing with humans and creating technology beyond our control. The problem is that many of these often fantastic claims, known as ‘doomerism’, are common currency in Silicon Valley. They are not subject to verification or peer review and it’s unclear whether these ‘extinction claims’ are based on real scientific evidence or being manipulated for competitive advantage. 

Nevertheless, there are real dangers which do require independent, enforceable safety regulation before models are released to the public, and ongoing monitoring to ensure compliance. This is essential if the tremendous potential benefits of AI are to be harnessed in the interests of humanity.

This won’t happen in the US given Trump’s stance. Big Tech has convinced the US Federal Government to adopt a light touch approach due to the “threat to national security” (China). AI companies have gone to market early to test and refine products, despite the possible risks. Under the White House’s executive directives, which OpenAI lobbied for, firms voluntarily submit advanced models to the government just before public release. Despite the hype, it is likely that the ‘Deceleration Pact’ will turn into another voluntary, self-regulation agreement with the government.  Whether it will be effective in preventing future harms is another matter.  

Other countries need to work together to try and mitigate these risks. In the UK, the regulatory framework is weak and based on the belief that an alliance with US Big Tech will deliver economic growth. This policy was pioneered by Peter Mandelson, who had a reputation for treating OpenAI’s Sam Altman as his “chief AI buddy.” While Ambassador to the US, Mandelson steered the UK away from EU-style safety regulations, towards a pro-innovation approach. In February 2025, Mandelson explicitly messaged Technology Secretary Peter Kyle, advising him to “benefit from more positive language about AI up front before you get into the security stuff.” This approach influenced the Starmer government, and it proceeded to sideline plans for a comprehensive AI Act and rejected the idea of mandatory legal ‘kill switches’ on advanced AI models, believing that pre-emptive legal bans could stifle economic growth.

Mandelson also helped broker the US-UK tech partnership pushing the line that it was essential for national security and defence and the UK should fast-track AI integration rather than holding it back through excessive regulation. When Trump visited the UK in September 2025, he signed a “world-leading Tech Prosperity Deal” with Starmer. This was soon followed by a Strategic Partnership between the spy-tech AI company Palantir and the MoD. Concerns have been expressed over the fact that Palantir was a client of Mandelson’s lobbying company Global Counsel. The impact of this deal is now becoming increasingly apparent, as Palantir’s controversial activities proliferate and AI hyperscale data centres are being developed across the country.

To address the threat of AI Frontier models causing harm, the government set up the AI Security Institute (AISI) – a research and evaluation body. However, it does not have regulatory powers, and companies are not legally required to submit their models. AISI operates via ‘goodwill’ agreements with AI tech labs. The fact that Anthropic recently declined to submit its latest ‘Mythos’ model to the AISI for pre-release review, exposes the limits of this voluntary approach. Even when tech companies cooperate and the AISI uncovers critical safety flaws, there is no power to stop the company from marketing the software.  An example occurred during the evaluation of OpenAI’s GPT-5.5. The AISI identified a universal ‘jailbreak’ vulnerability that bypassed the model’s cyber safeguards. OpenAI proceeded to release the model publicly before the vulnerability was resolved. 

The UK’s legal regulatory framework is highly fragmented across numerous bodies such as the Information Commissioner, Ofcom, the Financial Conduct Authority, and others. These focus on AI only once it has been deployed. Thus, the current legal setup places the burden of AI safety on the point of deployment (the businesses and public services using the tech) rather than the corporations designing the models. 

The lack of statutory oversight has led to urgent calls for reform. Parliament’s Joint Committee on Human Rights (JCHR) issued a highly critical report on 14th September warning that the UK’s fragmented regulatory framework is entirely unfit for purpose, demanding the government introduce a formal AI Bill.

The result of the Mandelson-Starmer policy is that US AI companies tend to view the UK as a friendly, malleable gateway to Europe, certainly compared with the EU, which is seen as an overly bureaucratic threat to their business models, with its commitment to strong regulatory enforcement, while simultaneously building its own AI infrastructure.

This is a critical time for the Burnham government. The safety crisis engulfing US Big Tech gives the UK the opportunity to carry out a fundamental review of its AI policy and its relationship with US Big Tech. The government has two months to respond to the JCHR report. A decision to abandon the voluntary approach and adopt a robust framework for AI regulation will be a crucial test of its willingness to change course. 

Effective AI safety regulation is urgent, but regulation should not focus exclusively on the possibility of future AI escapes and existential threats. There are very real AI-related threats that are happening in the world today, which the government also needs to consider:

It won’t be easy to move away from the Starmer-Mandelson approach. The government will need to make alliances across the world, including the Global South, the EU and Canada/Australia to develop an effective rules-based order under the auspices of the UN.  

There will be strong and sophisticated opposition from Big Tech lobbyists who helped ‘steer’ the policy in the first place. Palantir could already be gearing up for it by hiring Lord Tom Watson as a senior Vice President. Palantir’s chief executive Alex Karp said: “This is just the latest chapter of Tom’s forty-year fight for better public services. We are fortunate to have him guide us as we help the British government to deliver for the British public.”  

Tom joins the ranks of other notable, former UK politicians who have offered their services to Big Tech companies – Nick Clegg (META), Rishi Sunak (Anthropic and Microsoft), George Osborne (OpenAI). Sir Keir Starmer’s former AI advisor Matt Clifford has also joined Anthropic. 

Frank Hansen is a former Councillor in the London Borough of Brent.

Image:https://www.picpedia.org/chalkboard/a/artificial-intelligence.html License: Creative Commons 3 – CC BY-SA 3.0 Attribution: Alpha Stock Images – http://alphastockimages.com/ Original Author: Nick Youngson – link to – http://www.nyphotographic.com/ Original Image: https://www.picpedia.org/chalkboard/a/artificial-intelligence.html