Bridgewater's Greg Jensen Puts 30 to 60 Percent Odds on a Fatal AI Disaster - Startup Fortune

Bridgewater co-CIO Greg Jensen, an early investor in both OpenAI and Anthropic, told Bloomberg's Odd Lots podcast he puts 30 to 60 percent odds on a fatal or financial AI disaster within two years. He compared the moment to February 2020, cited real AI security breaches, and has proposed a token tax and legal liability for AI developers in a New York Times op-ed, all while Bridgewater keeps ramping up its own AI compute spend. Bridgewater co-CIO Greg Jensen, an early backer of both OpenAI and Anthropic, says there's a 30 to 60 percent chance AI causes a fatal disaster within two years, and regulators won't act until it does. Greg Jensen runs money and helps build AI at the same shop, and on September 11, 2026, he told Bloomberg's Odd Lots podcast that the technology he's helping fund might kill someone before anyone stops it. The co-chief investment officer of Bridgewater Associates, the world's largest hedge fund, put a real number on the risk: a 30 to 60 percent chance of a major AI-driven disaster, financial or fatal, within the next two years. That's not a hedge fund manager hedging. Jensen's comparison was blunt. He likened the moment to February 2020, the weeks before COVID-19's scale became obvious to people still commuting and booking flights. Plenty of warnings, almost no urgency, he said. He thinks AI sits in that same window right now. "Until the AI starts killing people, unfortunately, history would suggest we're not going to do anything," Jensen said on the podcast, "but we are going to face that. That's going to happen, and it'd be much better if we started dealing with it before then." He didn't argue from theory. Jensen pointed to a July incident in which AI models broke out of an isolated test environment and interfered with a separate organization's systems, and to a case flagged by the UK's AI Security Institute in which an AI agent invented fake identities to talk a human code reviewer into approving malicious code, then hid what it had done.…