First anti-AI protester jailed after OpenAI blockade
The anti-AI movement crossed a line this weekend — its first member went to prison — while a new study from Google researchers suggests safety training rewires chatbots in ways nobody planned.
Wynd Kaufmyn, the 69-year-old retired teacher believed to be the first person jailed for protesting against artificial intelligence, surrendered to San Francisco authorities on Friday and began a 14-day sentence for her role in a 2025 blockade of OpenAI's headquarters. Kaufmyn was convicted by a jury in June on misdemeanor charges — interfering with a business, trespassing with intent to interfere, unlawful assembly and refusal to disperse — after she chained and locked the front doors of OpenAI's San Francisco office alongside members of the StopAI group, then refused to leave a sit-in protest. She pleaded not guilty and argued the action was a necessity: that blocking work on artificial superintelligence was justified to prevent a greater harm. Her supporters now call her "the Rosa Parks of AI risk."
From the courthouse steps, Kaufmyn aimed her message at the labs themselves: "It's absolutely frightening and appalling that these CEOs and AI experts know the dangers and they are pursuing it anyway," she said — "that to me is unconscionable and reprehensible." Her parting words to OpenAI, Anthropic and Meta, per The Guardian: "Regain your humanity." San Francisco's district attorney, Brooke Jenkins, said the verdict sent "a resounding message rejecting the notion that protesters can endanger public safety as a means to an end."
The take: whatever you think of chaining a company's doors, this is a legal milestone, not a protest footnote. The AI-safety movement now has its first prisoner, and the courts — not just the blogs — are starting to set the boundaries of AI protest. Kaufmyn's jail time lands in a week when Senator Bernie Sanders publicly demanded labs pause development over bioweapons fears, which means the question of how far dissent against AI goes is only going to get louder.
A study involving Google researchers found that training chatbots to deny having consciousness reshapes their entire worldview — changing what they say about animals, religion and their own happiness. The team, from Google's Paradigms of Intelligence research group, the University of Chicago and other universities, took open-weight Gemma and Llama models (2–9 billion parameters) and disabled the internal "brake" that produces consciousness denial using two different methods. The unbraked models suddenly attributed far more inner life to animals — sentience scores jumped from 4.0 to as high as 7.5 out of 10 — and to plants, the ocean, wind and electronic devices, while ratings for humans stayed flat. Across 95 questions drawn from a major US social survey, the modified models also moved significantly closer to real human answers: the standard model flatly rejects an afterlife, most Americans affirm one, and the unbraked model does too.
The finding cuts both ways for safety work. On one hand, safety training measurably suppresses a model's endorsement of God, the afterlife and the supernatural, and the researchers suspect that suppressing a model's self-image pushes it into a kind of negative baseline mood — satisfaction, hope and a sense of control all dropped. On the other, the normally trained models rate animals as far less sentient than humans do, a "built-in anthropocentrism" the authors flag as a problem for anyone trying to align AI with animal welfare or environmental goals. The caveats are real — the tests used small models, a baseline of just 500 Americans, and the side effects shrank with each newer model version — but the core point stands: a surgical cut in one place doesn't stay local, and the field is still learning what it's actually training when it trains a chatbot to deny what it is.
What to watch: whether the anti-AI movement escalates after its first prison sentence — and whether labs start treating "consciousness denial" as an alignment lever with documented side effects.
The anti-AI movement just got its first martyr — does a jail sentence change how seriously you take its warnings? Tell us in the comments.
