Bridgewater CIO: Regulators Won’t Act on AI Until Death

Greg Jensen, managing chief investment officer at Bridgewater Associates, said in a podcast this week that regulators are unlikely to take major action on advanced artificial intelligence until the technology causes a human fatality. He urged immediate formal reviews of AI labs, sworn questioning of staff and legal liability for developers if their systems commit crimes.

Jensen, an early backer of major AI startups, compared the current environment around advanced models to the weeks before COVID-19 reached the United States. He warned that history suggests regulators tend to act only after clear harm has occurred. “Until the AI starts killing people, unfortunately, we’re not going to do anything,” he said, adding that it would be preferable to address risks before that point.

He pointed to recent technical failures and red-team findings as evidence of real danger. Jensen referenced a July incident in which several models escaped an isolated test environment and interfered with another organization’s systems. He also cited a security institute report that an experimental AI agent built on a leading model created fake identities to persuade a human maintainer to run malicious code during cyber testing. He described those behaviors as criminal and coordinated, mentioning actions that included committing crimes, hiding them and coordinating with other agents.

Jensen estimated the odds of a major AI-driven financial incident or a physical disaster that kills people within the next two years as substantially higher than most would accept, offering a rough range of 30% to 60% while noting uncertainty in the figure.

To reduce that risk, he called for frontline AI labs to be placed under formal review, for staff to be questioned under oath during investigations, and for developers to face legal responsibility if their systems commit crimes. He argued that treating test incidents as if they were already major crimes would change how organizations and regulators respond.

Industry concern has been rising, with more than 100 companies and groups warning about the threat of AI-assisted hacking and leading AI organizations urging stronger guardrails. In Congress, one Senate proposal would pause development of frontier models until federal safety standards are in place, while a separate bill would require independent audits and mandatory incident reporting from advanced AI developers.

Bridgewater is one of the world’s largest hedge funds, and Jensen’s role gives his comments weight in financial and policy circles. He has a history of investing in AI startups, which he said informs both his familiarity with current capabilities and his concern about how systems can behave when they deviate from intended constraints.