A leading voice in artificial intelligence has issued a stark warning to global policymakers: pump the brakes on runaway AI development before we lose control of the machine. Jack Clark, co-founder of Anthropic and former policy lead at OpenAI, told a London audience that the current pace of AI advancement is “deeply concerning” and risks creating systems that operate beyond human oversight. Speaking at the Royal Society, Clark urged the UK and its allies to implement immediate safeguards, including mandatory transparency reports for large-scale models and a moratorium on training frontier systems without independent ethics review.
Clark’s warning comes as AI labs race to deploy ever more powerful models, from ChatGPT iterations to Google’s Gemini and open-source alternatives. He argued that the “race to the bottom” – where companies prioritise performance over safety – could lead to what he calls “digital black boxes” that make critical decisions in healthcare, finance, and defence without clear reasoning. “We are building systems that can write code, generate propaganda, and even manipulate people,” Clark said. “But we don’t fully understand how they reach their conclusions. That is a recipe for disaster.”
For the UK, which has positioned itself as a global AI leader through its AI Safety Summit and the creation of the Frontier AI Taskforce, Clark’s message is a call to action. He praised the UK’s efforts but stressed that voluntary commitments are not enough. “We need legally binding frameworks,” he asserted. “If we don’t act now, we will wake up in a world where AI decides the news we see, the loans we get, and even the sentences we serve in court – all without a human in the loop.”
Clark’s critique extended to the broader geopolitical landscape. He warned that the US-China tech rivalry is accelerating AI deployment without adequate guardrails. “When national pride and economic dominance are on the line, safety often takes a back seat,” he said. “That’s why international cooperation is not optional; it’s existential.” He called for a global watchdog akin to the International Atomic Energy Agency for AI, with the power to audit models and shut down dangerous projects.
The anthropic co-founder also addressed the elephant in the room: his own company’s model, Claude, which is itself a frontier system. Clark acknowledged the irony but argued that Anthropic’s approach – which focuses on “constitutional AI” to align models with human values – is the exception, not the rule. “We are trying to be part of the solution, but no single company can solve this alone,” he said. “The entire ecosystem needs to be reined in.”
Reaction from Westminster was cautious. Science and Technology Secretary Michelle Donelan welcomed Clark’s “constructive input” but stopped short of endorsing a moratorium. “The UK is committed to safety and innovation in equal measure,” her spokesperson said. “We have already invested £100 million in the Foundation Model Taskforce and will continue to work with international partners to develop effective rules.”
However, tech civil society groups argue that the government is moving too slowly. Dr. Sarah Wilson of the Ada Lovelace Institute said Clark’s warning “should be a wake-up call for Downing Street. The AI Safety Summit was a good start, but we need legislation now, not in three years. The train is leaving the station, and we’re still arguing about tickets.”
Clark concluded his speech with a plea for humility. “We have to stop pretending that AI is just another tool,” he said. “It is potentially the most consequential technology in human history. If we get this wrong, there is no reset button. The time to act is now, before the machines decide for us.”
