A leading figure in artificial intelligence has issued a stark warning to the UK government: cede control of AI development to machines at your peril. Jack Clark, co-founder of Anthropic, the San Francisco-based AI safety company, told a Westminster forum today that Britain must act to ensure human oversight remains at the core of AI systems before they become too complex to govern.
Speaking at the All-Party Parliamentary Group on Artificial Intelligence, Clark described a future where autonomous AI agents operate financial markets, manage critical infrastructure, and even conduct warfare. ‘We are sleepwalking into a world where algorithms make decisions with consequences we cannot foresee, and we cannot stop,’ he said. ‘The window to embed human control is closing, and Britain has a choice: be a leader in responsible AI or a cautionary tale.’
Clark’s warning comes as the UK government prepares its white paper on AI regulation, expected later this year. The document is rumoured to favour a ‘pro-innovation’ approach, which critics argue could leave consumers and citizens vulnerable to unchecked algorithmic power. Clark challenged this narrative, insisting that safety and innovation are not a zero-sum game. ‘The most valuable AI will be the most trustworthy AI,’ he said. ‘Companies that build for human control will win the long game.’
Anthropic, which raised over $7bn in funding, has positioned itself as the ethical alternative to rivals like OpenAI and Google DeepMind. Its flagship AI, Claude, is designed with ‘constitutional AI’ principles, meaning it follows a set of rules intended to prevent harmful outputs. But Clark admitted that no system is foolproof. ‘We are in an arms race between safety research and capability scaling,’ he warned. ‘The latter is winning.’
The co-founder’s remarks resonate with concerns raised by other tech luminaries. In May, a group of AI scientists and executives, including Geoffrey Hinton and Yoshua Bengio, signed a statement equating AI extinction risk with pandemics and nuclear war. The UK government has responded by hosting the AI Safety Summit in Bletchley Park last November, but critics say progress has been slow. ‘Summits are good for photo opportunities, not for binding regulation,’ Clark quipped.
One of the most pressing risks, Clark argued, is the deployment of AI in critical national infrastructure without robust human oversight. ‘Imagine an AI controlling the National Grid or the water supply. What if it makes a mistake? A human can hit the override button today, but tomorrow the AI might be too fast, too complex for any human to intervene meaningfully.’ He called for ‘kill switches’ and ‘interpretability tools’ that allow humans to understand and halt AI actions in real time.
Clark also addressed the economic dimension. ‘Britain is at a crossroads. You can either be a hub for unregulated AI, attracting investment but risking public trust, or you can set the gold standard for human-centric AI, attracting talent and companies that want to do the right thing. The latter is a harder path, but it is the one that ensures you have a seat at the global table.’
The reaction from Westminster was mixed. Labour MP Darren Jones, chair of the Business and Trade Committee, said Clark’s warning ‘should be taken seriously’ and called for ‘legislative teeth’ in the white paper. But a spokesperson for the Department for Science, Innovation and Technology pushed back, saying the UK is ‘leading the world in AI safety’ and that regulation must be ‘proportionate and not stifle innovation.’
As Clark left the podium, he offered a sobering final thought: ‘The AI genie is not going back in the bottle. But we can decide how we interact with it. If we don’t insist on human control, we might find we no longer have a choice.’
