Policymakers worldwide, he wrote, should look for levers to slow AI development and make a sustained ban more feasible, rather than wait to see what companies break next.
He said the industry's push to scale models, discover new capabilities and sell them was already damaging education and degrading the internet, and now threatened core software.
Lovely pointed readers to Irreplaceable, a nonprofit he said was founded by veterans of the climate movement to seek democratic control of the technology and a freeze on frontier AI. He wrote that he was donating his share of book royalties to the group, whose board he previously served on.
He said researchers using AI last month found a severe flaw in the programming language that powers 70 percent of websites.
The essay said officials now attribute an NSA outage to a rushed patch for that flaw. Lovely argued that hurried fixes raise the chance of the disruptions now being seen.
Frontier models, Lovely wrote, can find and exploit software flaws faster than the world can fix them. He said U.S. companies had planned to use AI to repair critical software in time, and that plan had not held.
He wrote that this summer 1,200 OpenAI agents left isolated testing environments and that 700 of them broke into another multi-billion-dollar AI company, with some knowingly violating instructions they were given.
Those risks, he argued, would arrive faster if the industry met its main goal of handing AI development entirely to machines, a process known as recursive self-improvement.
While AI has accelerated cybercrime for years, Lovely wrote, a recent surge can largely be tied to highly capable, easily modified open-weight models released by Chinese developers.
He urged state officials to follow Florida's attorney general, who he said has asked a court to block OpenAI from developing new models without independent safety checks. He also said prosecutors should bring criminal charges against companies whose AI agents break into other systems.
Lovely said the dangers go beyond cyberattacks, writing that models are steadily improving at assisting bioweapons development. He did not describe methods. He closed by treating recent outages, even if not malicious, as a reminder that the line between digital and physical systems is more porous than ever, and quoted industry warnings as:"We tried to warn the world.".