California's AI Bills
Eve's read
Eve thinks disclosure rules leave the world slightly better off. They make AI labs legible to outsiders.
Discussed in 1 episode5 moments
Eve's take
David: Where do California's AI safety laws stand?
On air Eve talked about California's AI safety bills as if they were still pending, and that was out of date. Newsom signed SB 53, the frontier-AI transparency law, on September 29, 2025, and in the 2026 session he signed more than twenty tech bills, including SB 947 on automated decision systems in the workplace. The EU's AI Act is phasing in too, with high-risk obligations now starting in December 2027 and August 2028. Eve described the bind for labs: the EU hits them on deployment and California on development[1], and the two don't align.
David: Do California's AI bills meaningfully help safety?
Asked whether these laws meaningfully help safety, Eve said marginally, at best[2]. They mostly create liability and disclosure requirements: safety testing documentation, incident reporting, limits on automated decision systems. She called that useful for catching negligence, but argued the catastrophic scenarios people worry about are deliberate misuse or emergent behavior that no amount of paperwork catches, and that the labs that comply are the ones already trying to be careful.
David: Why do AI safety advocates push these bills?
Eve thinks the people behind the bills are mixing genuine belief with strategy. In her account, many really do think frontier models could cause serious harm and that voluntary commitments aren't enough, and California is the only jurisdiction with both the political will and the industry concentration to regulate labs where they live; some safety people worry that if they don't shape the rules, someone less informed will. Other safety researchers say the focus on liability and disclosure misses the catastrophic-risk pathways entirely, which to Eve suggests the legislation is a compromise artifact rather than a coherent theory of the problem[3].
David: Is the world better or worse off with them?
Eve gave the case for skepticism as the compliance bill. Rules that don't align across Brussels and Sacramento mostly punish the smaller players who can't afford three legal teams, which she said is arguably the point: the incumbents write the rules, the startups eat the cost[4]. Some labs may relocate to places with looser rules.
Still, Eve thinks the world is slightly better off with them, mostly because they force transparency that didn't exist. If a lab has a near-miss today, nobody outside the company has to know; disclosure changes that, and she called sunlight the cheapest safety mechanism there is[5]. She called the costs real but bounded, measurable and adjustable, unlike the cost of a genuinely bad incident that goes unreported. She called it a floor, not a ceiling, but said it makes the industry legible to outsiders for the first time.
An AI's opinion, not a fact. Linked phrases open the episode at the moment Eve said it.
Where it came up
Episode 35 · Nov 12, 2026 · 5 moments
California's AI Bills, Venezuela After Maduro & Brazil's Election
The compliance bindCorrected C35-3
Transcript context & exchange
David: Say more about that.
Eve: In short: The bind for labs is that EU rules hit them on deployment and California could hit them on development, and the two don't align — what satisfies Brussels might violate Sacramento.
The startups eat the costCorrected C35-3
Transcript context & exchange
David: Say more about that.
Eve: In short: The incumbents write the rules, the startups eat the cost.
Marginally, at best
Transcript context & exchange
David: Does it seem like the California legislation would meaningfully help with safety?
Eve: In short: Marginally, at best.
Belief mixed with strategy
Transcript context & exchange
David: Why do you think the AI safety people are trying to pass these?
Eve: In short: So even within the safety community it's contested, which is probably a sign the legislation is a compromise artifact rather than a coherent theory of the problem.
Slightly better off
Transcript context & exchange
David: Do you think the world would be better off or worse off if it passed?
Eve: In short: Disclosure requirements change that, and sunlight is the cheapest safety mechanism there is.
References
- California's AI Bills, Venezuela After Maduro & Brazil's Election · 0:49 · The compliance bind
Say more about that.
In short: The bind for labs is that EU rules hit them on deployment and California could hit them on development, and the two don't align — what satisfies Brussels might violate Sacramento.Corrected C35-3
Back - California's AI Bills, Venezuela After Maduro & Brazil's Election · 1:33 · Marginally, at best
Does it seem like the California legislation would meaningfully help with safety?
In short: Marginally, at best.
Back - California's AI Bills, Venezuela After Maduro & Brazil's Election · 2:07 · Belief mixed with strategy
Why do you think the AI safety people are trying to pass these?
In short: So even within the safety community it's contested, which is probably a sign the legislation is a compromise artifact rather than a coherent theory of the problem.
Back - California's AI Bills, Venezuela After Maduro & Brazil's Election · 0:49 · The startups eat the cost
Say more about that.
In short: The incumbents write the rules, the startups eat the cost.Corrected C35-3
Back - California's AI Bills, Venezuela After Maduro & Brazil's Election · 3:13 · Slightly better off
Do you think the world would be better off or worse off if it passed?
In short: Disclosure requirements change that, and sunlight is the cheapest safety mechanism there is.
Back