California’s SB 1047 bill, aimed at preventing AI disasters, has encountered opposition from various Silicon Valley parties. In response, California lawmakers have made amendments to the bill, incorporating suggestions from AI firm Anthropic and other opponents.
The bill recently passed through California’s Appropriations Committee with significant changes, moving closer to becoming law, according to Senator Wiener’s office as reported by TechCrunch.
Senator Wiener expressed, “We have accepted reasonable amendments that address the concerns raised by Anthropic and the industry. These changes, along with previous modifications to accommodate the open-source community, aim to foster innovation,” in a statement to TechCrunch.
While SB 1047 still intends to hold developers accountable for preventing AI-related disasters, it now lessens the government’s power to regulate AI labs.
Key Amendments to SB 1047:
The bill no longer permits California’s attorney general to sue AI companies for safety negligence before a catastrophic event occurs, as suggested by Anthropic. Instead, the attorney general can seek injunctive relief and can only sue if an AI model causes a catastrophic event.
The Frontier Model Division (FMD) was removed from the bill. However, the Board of Frontier Models is retained within the Government Operations Agency, now comprising nine members. The board will set compute thresholds, provide safety guidance, and establish regulations for auditors.
AI labs are no longer required to submit safety test results under penalty of perjury. Instead, they must provide public statements outlining safety practices without facing criminal liability.
The bill now requires developers to exercise “reasonable care” to ensure AI models don’t pose significant risks, replacing the previous requirement of “reasonable assurance.”
An exemption for fine-tuned open-source models spending less than $10 million is included, with primary liability remaining with the original developer.
Reasons for Amendments:
Despite facing opposition, SB 1047 has navigated through the legislature. These changes are expected to address concerns and offer a less controversial bill for Governor Newsom to sign.
Governor Newsom, known for supporting AI innovation in California, has not yet commented on SB 1047 publicly.
However, staunch critics of SB 1047 remain unsatisfied with the amendments, as the bill still holds developers accountable for AI dangers, a point of contention not universally accepted.
Following the bill’s passage, California congress members penned a letter urging Governor Newsom to veto SB 1047, citing concerns about its impact on the state’s startup community and scientific development.
Next Steps:
SB 1047 will undergo a final vote on the California Assembly floor. If approved, it will return to the Senate for another vote due to recent changes. Ultimately, Governor Newsom will decide whether to veto or enact the bill into law.
