In a stark address to policymakers in London, the co-founder of Anthropic, the artificial intelligence firm behind the Claude chatbot, declared that AI systems must be designed to serve Britain’s national interests. Speaking at a closed-door session on AI governance, the executive emphasised that no algorithm should operate beyond human control—especially in matters touching national security.
The speech came amid a flurry of government consultations on the UK’s AI Safety Bill, which seeks to regulate frontier models. The Anthropic co-founder, a former OpenAI researcher, argued that the pace of development has outstripped the ability of existing institutions to keep society safe. “We are building systems that can synthesise biological weapons, manipulate financial markets, or conduct disinformation campaigns at scale,” he told the room. “The only check on that power is a human in the loop, empowered to say no.”
His warning resonates deeply with the government’s own security assessments, which have flagged risks of adversarial states weaponising general-purpose AI. The National Cyber Security Centre recently warned that large language models could lower the barrier to cyberattacks, enabling less skilled actors to craft sophisticated phishing campaigns or exploit zero-day vulnerabilities.
But the Anthropic co-founder insisted that regulation alone is not enough. He called for a “digital sovereignty” framework where critical AI infrastructure is hosted onshore and subject to British legal oversight. “We cannot outsource our safety to Silicon Valley boardrooms, no matter how well-intentioned,” he said. “If an AI system makes a decision that affects a British citizen, that decision must be traceable to a human accountable under British law.”
The intervention comes as the Prime Minister tries to position Britain as a global hub for AI safety research, balancing economic opportunity with existential caution. The government has already invested £100 million into a new Foundation Model Taskforce, but critics argue that the private sector still holds too much sway over research directions.
Anthropic itself has been a vocal advocate for “constitutional AI” a technique that uses a set of guiding principles to align models with human values. Yet the co-founder admitted that no technical fix can replace democratic oversight. “We can train models to be polite, but we cannot train them to be loyal to a nation’s values without embedding those values in the procurement process,” he said.
The speech concluded with a call for Britain to lead on AI ethics, not just safety. “Safety is about preventing harm. Ethics is about deciding what kind of future we want. Britain has the chance to write the rulebook for the 21st century, but only if it demands that AI serves the people first—not the shareholders.”
