SB 1047
California Senate Bill 1047, proposed legislation requiring safety evaluation and risk mitigation for the most powerful frontier AI models trained above a defined compute and cost threshold.
As defined in 379 - Regulating Artificial Intelligence
Where it comes up · 2
- 379 - Regulating Artificial Intelligence7 claims
- Anthropic CEO Dario Amodei Talks Scaling Laws, AI Arms Races, and Radical Abundance3 claims
Most important claims · 10 of 10Search all claims for “SB 1047” →
- This is not a regulatory-capture bill: in classic regulatory capture incumbents design rules that only they can satisfy and that lock out competitors, but here only large labs are subject to regulation and startups are not—and the fact that Meta and Google oppose the bill shows it is the opposite of capture, since captured incumbents would support it.379 - Regulating Artificial Intelligence
- Critics cannot consistently claim AI risks are pure science fiction while fighting hard against the bill, because if the harms will never happen the bill would never be triggered and would cover nothing—so the intensity of opposition (led by a16z) contradicts the claim that the risks are fabricated.379 - Regulating Artificial Intelligence
- Under the bill, if a lab performs the required safety evaluations it is not liable even if its model is later misused, only the attorney general can file suit and only for non-compliance, and broader tort liability for AI harms already exists today under existing California law—so the liability fears are overstated and claims of prison for developers are false.379 - Regulating Artificial Intelligence
- Because the bill prescribes no particular evaluation method and leaves it to companies and to national, international, and academic standards bodies to develop progressively better risk evaluations, it incentivizes more safety homework and stimulates the creation of systems that are both capable and safe.379 - Regulating Artificial Intelligence
- The bill is not contingent on training, releasing, or being located in California—it applies to any covered lab doing business in California's economy, so the claim that labs or startups will flee California to avoid it is unfounded since none will plausibly forgo the world's fifth-largest economy.379 - Regulating Artificial Intelligence
- Most open-source AI from academia and startups is much smaller than the $100M/10^26 FLOP threshold because they lack the resources to train at that scale, so that activity continues unaffected; the bill treats open and closed source equally but removes shutdown obligations once a model leaves your possession and exempts sufficiently fine-tuned derivative models.379 - Regulating Artificial Intelligence
- SB 1047 requires that anyone training and releasing a model exceeding 10^26 FLOP and costing at least $100 million perform reasonable pre-release safety evaluations and, if those evaluations show significant risk of catastrophic harm, take reasonable steps to mitigate it—getting ahead of risk rather than waiting for catastrophe.379 - Regulating Artificial Intelligence
- SB 1047's original version with pre-armed enforcement (government establishing what tests must be run) is problematic because tests for catastrophic AI scenarios are very new and almost none of the relevant catastrophes have occurred, making government-mandated tests potentially counterproductive or politically misused.Anthropic CEO Dario Amodei Talks Scaling Laws, AI Arms Races, and Radical Abundance
- Moving AI company headquarters out of California in response to SB 1047 would be 'theater' and 'negotiating leverage' with no substantive impact, because the bill applies to companies doing business in California or deploying models in California, not to incorporation jurisdiction.Anthropic CEO Dario Amodei Talks Scaling Laws, AI Arms Races, and Radical Abundance
- California's SB 1047 AI regulation was initially too heavy-handed but addressed roughly 60% of Anthropic's concerns in its revised form, and Anthropic became substantially more supportive of the amended bill while maintaining an analytical stance rather than partisan advocacy.Anthropic CEO Dario Amodei Talks Scaling Laws, AI Arms Races, and Radical Abundance