While speaking to BBC Newsnight, Clark said the AI industry currently has a “gas pedal” but no “brake pedal” and argued that governments and policymakers lack necessary mechanisms to ensure humans remain in control of systems that grow ever more powerful.

This comes as Anthropic is widely viewed as the frontrunner in the competition between major AI firms, growing rapidly and preparing for its IPO debut. Anthropic has positioned itself as the more safety-focused AI company, regularly warning that advanced models could create serious social, economic, and security disruptions if not properly governed.

Clark went on to reveal that around 80% of Claude’s code was written by itself and suggested that reaching 100% could be achieved within two years. If AI can increasingly build and improve its own infrastructure, the question of human oversight becomes far more urgent in the face of exponential improvement.

Clark’s concerns are not only technical; he also pointed to potential disruptions across the economy from autonomous AI agents carrying out routine tasks. Major tech companies have already fired thousands of workers due to the growing ability of AI to do their jobs. In his view, AI now needs a proper policy framework and regulatory structure to maintain public trust and allow society to benefit from the technology without depending solely on the judgment of company leaders.

However, Clark was not entirely pessimistic. He argued that creativity, curiosity, and broad thinking may become even more valuable in an AI-driven economy because, while AI is great at execution, there is relatively little evidence that these systems actually generate original ideas in the same way humans can.

Overall, the broader takeaway message is clear: AI’s potential benefits are enormous, but so are its risks, and the industry cannot rely on speed alone. As systems become more capable, we must be vigilant in ensuring our regulation, oversight, and accountability frameworks evolve to match them. This is something that must be pursued as enthusiastically as the race to improve AI, for it is only a matter of time before the technology evolves beyond the structures designed to govern it.

“You want the option to be able to take your foot off the gas and put your foot on the brake. Right now, it’s like the AI industry has a gas pedal, but it doesn’t have a brake pedal.” Jack Clark, Co-Founder of Anthropic