Over 400 people gathered outside OpenAI’s Mission Bay headquarters in San Francisco on a Saturday morning in July. They were carrying handmade posters, a marching band, a Barney costume, and a white van that was selling slushies. “Shame!” was chanted outside Andreessen Horowitz. They created a commotion outside Google DeepMind’s Rincon Park offices and Anthropic on Howard Street. After that, they returned home. And the businesses continued to expand.
In some ways, the entire narrative of the current anti-AI labor movement revolves around the discrepancy between the protest’s intensity and the industry’s speed. People traveled ten hours from Los Angeles and two hours from Sacramento to participate in the “Stop the AI Race” march on July 11, 2026, which was the coalition’s second significant mobilization in a single year. They demanded that the CEOs of the most powerful AI labs in the world publicly pledge to halt the development of frontier models. Hunter Glenn, one of the protesters, lost his job as a copywriter when his employer started using AI. Another was an AI researcher who, fearing surveillance, refused to reveal his real name. They were both there because they believed the industry was heading too quickly in the direction of something it didn’t completely comprehend.
The movement’s main demand is purposefully limited: CEOs of all big AI labs should publicly pledge to halt frontier model training, provided that all other significant laboratories globally do the same in a credible manner. Depending on how optimistic you are about international coordination, it can be either pragmatic or somewhat symbolic because it is a conditional freeze rather than a permanent prohibition. When skeptics bring up China, the organizers point out that any deal would have to include international laboratories. However, the first step is to record American CEOs. So far, none have accepted it.
The parallel movement taking place within the firms themselves is what has given the street protests more weight than their headcount might indicate. In May, a petition opposing Meta’s Model Capability Initiative—a program that gathers employee computer usage data to train AI models—was signed by over 1,600 employees. Although the petition did not result in the program’s termination, it did cause an unusually public internal conflict for a business that normally keeps its employee complaints to itself.
In the same month, Google DeepMind employees in the UK decided to form a union, largely due to their disapproval of the company’s use of AI in military applications. In a combined letter, more than 600 laid-off Oracle workers claimed they had been trained AI systems prior to being fired. They are not demonstrators on the streets. The same questions are being raised by those who were inside the machine.
Beneath all of this lie substantial and likely underestimated labor numbers. Since 2025, about 400,000 IT workers have been laid off; in 2026 alone, over 150,000 of them left, with many of them specifically citing automation. According to a 2026 study by Stanford’s Institute for Human-Centered AI, a third of businesses now anticipate that AI will result in a reduction in their overall workforce, with software engineering and service operations anticipating the largest reductions.
In Silicon Valley, the engineers who were promised a career-long skill premium are suddenly witnessing entry-level coding roles vanish due to the very tools they helped create. San Francisco product designer Jenny Lin used a precision that seemed real rather than spectacular to convey the atmosphere of the July march: “AI is fast at execution, but execution was never the hard part of the job.”
