How a protest works before it wins, and AI workers' political money

A firsthand account of the Stop the AI Race march in San Francisco and the ways a protest can make progress without winning its demand, paired with an argument that political donations from AI company employees signal real unease about the industry.

Hundreds of people marched in San Francisco on July 11, stopping at OpenAI, Anthropic and Google DeepMind, with a single demand: that every major AI lab CEO publicly promise to pause frontier development if all the other major labs credibly do the same (Haven Harms)
No CEO has agreed so far, but the author argues protests also succeed by building a lasting coalition, spreading the message through coverage and social media, and making previously fringe positions look legitimate, which in turn makes narrower rules like safety testing seem moderate (Haven Harms)
The crowd went well beyond the AI safety world, mixing people worried about jobs, surveillance, the environment, extinction risk and their children's futures; organizers plan weekday protests, the next one on July 20, and a separate march in Washington is collecting pledges (Haven Harms)
A San Francisco Standard report finds employees at AI companies donating to candidates and super PACs that favor stronger regulation of their own industry, with roughly 59 of every 1,000 Anthropic staff giving to a federal campaign this cycle, far above earlier tech workforces (Mitch)
Mitch reads those donations as sincere misgivings rather than a strategy to capture regulators, but doubts that trying to do good from inside the labs can work, citing a former Google DeepMind researcher on how companies cross the red lines they set for themselves (Mitch)