This article was AI-generated as part of an experimental historical-content project. The date reflects the period being analyzed rather than the date the article was originally written.
On Sunday, California Governor Gavin Newsom vetoed SB 1047, the bill that would have required developers of the largest AI models to put safeguards and policies in place to prevent catastrophic harm. It was one of the most closely watched AI bills in the United States this year, and the debate around it divided AI companies, researchers and investors in unusually public ways.
I want to leave aside whether the veto was right. A different question will reach communications teams sooner, at AI developers and at the many companies now deploying AI: without a statute defining it, who gets to say what “responsible AI” means?
What the veto message actually argues
Newsom’s message is worth reading in full, because it does not dismiss the risks. It says “we cannot afford to wait for a major catastrophe to occur before taking action” and that “safety protocols must be adopted.” His objection is to the bill’s approach. He argues that focusing only on the most expensive, large-scale models “could give the public a false sense of security,” and that the bill does not consider whether a system is deployed in high-risk settings or uses sensitive data. He also notes that he has signed more than a dozen bills on specific, known AI risks in the last 30 days.
The bill’s author, State Senator Scott Wiener, called the veto “a setback” and argued that voluntary commitments from industry “are not enforceable.”
Both sides are claiming the same ground. Each describes its position as the responsible one.
A phrase without a definition
That is the reputation issue. “Responsible AI” now appears in product launches, safety pages, investor materials and job postings. Most of the time it refers to voluntary frameworks: internal policies, red-teaming, safety commitments, model documentation. A binding California standard for the largest models would have given part of that phrase a legal anchor. For now, California has none.
When a term has no external definition, it gets defined by whoever publishes the most about it and whoever is cited most often. Search results for “responsible AI” are full of company pages. An AI assistant asked what responsible AI means will summarize those pages. The meaning drifts toward the language of the organizations with the largest publishing budgets.
That is not sinister, but it is fragile. A claim that rests only on a company’s own description can be reframed overnight by an incident, a leaked document or a critic with a better headline.
What a careful company does with that
For communications and policy teams, a few things follow.
Be specific. “We are committed to responsible AI” is unfalsifiable, and readers increasingly skip past it. “We test models against these categories before release and publish the results here” can be checked, and is more likely to be quoted accurately.
Separate what is voluntary from what is required. Journalists, customers and eventually regulators will ask. Clear language now avoids looking evasive later.
Expect the frame to change. The veto message says California will keep working on legislation and points to federal work at the U.S. AI Safety Institute. A statement written for this moment should still hold up if a binding standard arrives.
The veto settled one bill. It did not settle who defines responsibility. Until the law does, the companies whose claims can be verified will have the strongest hold on the phrase.