Elon Musk’s company, xAI, claims that a state law violates the First Amendment right to withhold speech by compelling transparency over how it trains its models.

Ask an AI chatbot a question, and it answers with total confidence — even when it’s wrong. Chatbots have invented court cases, medical dosages and statements no one ever said. The companies that built these tools know exactly what data produced that confidence, but you don’t. That gap is the fight now playing out in courtrooms nationwide and in California.

Elon Musk is at its center. This year his company, xAI, sued California over a law demanding that artificial intelligence companies disclose how they train their models. Musk’s company claims the law violates its First Amendment right to withhold speech by compelling transparency.

But California’s law is not an outlier. Assembly Bill 2013, the Generative Artificial Intelligence Training Data Transparency Act, is part of a nationwide wave of legislation — all 50 states introduced AI-related bills, and 38 adopted roughly 100 measures — aimed at prying open the black box of technology now embedded in daily life. It requires AI companies to publish a high-level summary of their training datasets: Is it copyrighted? Was it modified? When was it collected? None of the law’s roughly dozen factual categories touches on proprietary methods or requires opinions from the companies; it simply gives researchers, users and journalists enough facts to judge these datasets for themselves.

Article continues below this ad

Anthropic and OpenAI have already complied with AB2013’s requirement to provide short public summaries. xAI sued instead,........

© San Francisco Chronicle