Greg Jensen, co-chief investment officer at Bridgewater Associates, delivered one of the bleakest assessments of AI regulation to date on September 11. His core argument: history suggests governments won’t act until something truly terrible happens.
“Until the AI starts killing people, unfortunately, history would suggest we’re not going to do anything,” Jensen said, comparing the current moment to February 2020, when COVID-19 was already spreading globally but most governments hadn’t yet mobilized a response.
A week of escalating warnings
Jensen’s remarks capped a stretch of days that saw multiple high-profile figures voice serious concerns about the trajectory of AI development. On September 9 and 10, researchers at Anthropic, including Jacob Coxon and Evan Hubinger, issued stark cautions about the current pace of AI research.
Coxon resigned from Anthropic during that window, citing the very risks he’d been studying from the inside. Both researchers warned that the field’s current trajectory could produce self-improving superintelligences, the kind of systems that don’t just follow instructions but rewrite their own capabilities.
Their estimate was chilling: a greater than 10% probability of human extinction from AI within the next decade.
Jensen suggested a major AI-related incident could occur within the next two years if no proactive measures are taken.
Washington starts paying attention
US lawmakers, including Rep. Lori Trahan and Sen. Ted Cruz, pushed for legislative action on September 9 and 10, calling for frameworks that would mandate transparency from AI developers and establish emergency pause mechanisms for high-risk models.
These proposals build on work done by a UN independent panel, which released a report in July flagging the rapid growth of AI capabilities. The panel’s central finding was uncomfortable: AI is advancing faster than both scientific understanding and government policy can keep up.
The UN panel specifically warned about the inadvertent development of deceptive or agentic systems, AI that can pursue goals autonomously and potentially mislead the humans overseeing it.