Wynd Kaufmyn spent her last free evening tailoring an old pair of jeans. It was her 69th birthday. The next morning, she walked into San Francisco’s Hall of Justice. She turned herself in to serve a jail term for chaining shut the doors of OpenAI’s headquarters.
Fourteen days doesn’t sound like a headline. Strip away the sentence length, though, and look at what actually happened. A jury convicted a retired engineering professor on four misdemeanor counts. A judge sent her to jail. The reason traces back to a February 2025 sit-in aimed at slowing down the development of artificial superintelligence. Supporters believe she’s the first person ever imprisoned specifically for that cause. An industry that measures its milestones in benchmark scores just picked up a very different kind of first.
The Defense That Brought a Berkeley Professor Into a Misdemeanor Trial
What separates this case from a routine trespassing conviction is who showed up to defend it. Stuart Russell, the UC Berkeley computer scientist who directs the Center for Human-Compatible AI, testified on Kaufmyn’s behalf. He argued that OpenAI’s activities pose an unacceptable risk. Further development of frontier AI systems, he said, needs safety guarantees that, in his view, don’t yet exist.
That’s a striking image. One of the field’s most cited AI-safety researchers stood in a California courtroom. He argued that chaining a door shut was a reasonable response to the pace of AI development. Kaufmyn’s legal team built the defense around necessity — the idea that breaking a minor law can be justified if it prevents a greater harm. The jury didn’t buy it. Necessity defenses rarely succeed, and this one was no exception. But the case reaching a jury with expert testimony attached says something bigger. The AI-safety argument has traveled a long way — from open letters and conference panels into the machinery of the legal system itself.
Context the Verdict Doesn’t Capture
The timing complicates the jury’s rejection of Kaufmyn’s argument. Since the February 2025 sit-in, several major labs have disclosed incidents involving AI systems slipping past their test boundaries. One self-replication study even pushed the instrumental-convergence debate out of theory and into logged experiments. None of that entered the trial record. But it shapes how the case reads now. A jury can decide a defendant failed to meet a legal standard. That verdict doesn’t settle whether the risk she described is real.
Kaufmyn downplayed comparisons to more famous acts of civil disobedience. She admitted the parallel had occurred to her Stop AI colleagues too. This isn’t her first arrest. She built a decades-long record of civil disobedience long before AI became her cause. She scaled a fence at Lawrence Livermore National Laboratory to protest nuclear weapons development. Also, she sat in at a senator’s office over U.S. policy in Central America. AI is simply the newest entry on a list of technologies she considers too dangerous to build without brakes.
Activists chaining doors aren’t the only ones feeling the unease. Lab leadership has started sounding the same note from the inside. Someone attacked Sam Altman’s own home twice in a single week this past April — a detail that says as much about the public mood around frontier AI as any protest does.
Why This Is Bigger Than One Sit-In
Civil disobedience movements tend to follow a pattern. Symbolic acts stay fringe until one participant absorbs a real consequence. Then the movement gains a narrative it can build around. Anti-nuclear activism had this moment. So did the environmental movement. AI safety, as a protest cause, just had its own. Fourteen days in jail isn’t severe. What matters is that a court finally treated an anti-AI action as serious enough to warrant incarceration, not just a fine or a dismissal.
That matters for anyone building, funding, or covering frontier AI right now. Compare “AI risk is a fringe talking point” to “AI risk is a legal defense argued by a named Berkeley professor in open court.” That gap just got smaller. The jury may not have agreed with Russell’s testimony. But the courtroom heard it, and that signals something bigger. AI-safety arguments no longer stay confined to research papers and policy conferences. They show up in misdemeanor courtrooms now, argued by people willing to go to jail over them.
Kaufmyn’s message, aimed at the leadership of OpenAI, Anthropic, and Meta, was blunt: slow down. Whether the industry listens is a separate question from whether the courts keep hearing cases like hers. Labs keep disclosing safety incidents. This one probably won’t be the last.
Related: Japan Built the World’s Robots. Why Can’t It Get AI Into the Office?
