CALI AI DISASTER BILL REVAMPED: TECH GIANTS SCORE WIN WITH MODIFIED LIABILITY, PROTECTION MEASURES

In a significant move that could shape the future of artificial intelligence (AI) in the United States, SB 1047, a bill designed to prevent AI disasters, has passed through California's Appropriations Committee, albeit with several key amendments proposed by various opposition groups, including AI firm Anthropic.

Previously, the legislation had allowed for the attorney general to sue AI firms for negligent safety practices before an incident took place. However, the revised bill ensures that the attorney general can only seek injunctive relief and sue following a catastrophic event linked to an AI model—thus changing the landscape for legal recourse against potential AI disasters.

One of the most distinctive changes in the bill is the omission of the Frontier Model Division (FMD), a body initially outlined for creation. Instead, the revised bill orders the establishment of the Board of Frontier Models under the existing Government Operations Agency. Responsibilities of this board include setting computing thresholds for covered models, providing safety guidance, and creating regulations for auditors.

The amendments have also lessened the onus on AI labs to confirm safety. Rather than requiring laboratories to submit certifications of safety test results under penalty of perjury, these entities are now only needed to make public statements outlining their safety practices.

Moreover, the wording within the bill concerning AI developers has become lenient too, suggesting they provide "reasonable care," not "reasonable assurance" that their models do not present significant risks.

Another shift within the bill aims to protect open-source fine-tuned models. Developers who invest under $10 million for fine-tuning a model will be exempt from developer categorization per SB 1047. Instead, responsibility will remain with the original developer.

Despite the amendments, resistance to SB 1047 persists amongst many congress members and industry figures. Critics argue that the bill fails to address main points of dispute and could stifle growth and innovation within the AI industry. The opposition has urged California Governor Gavin Newsom to veto the bill outright.

Now moving to California's Assembly floor for a final vote, if the bill is successful, it will return to the Senate for another vote due to the amendments. Only if it passes these hurdles will it be placed on Governor Newsom’s desk for the final decision; to either veto or sign into law.

The outcome of this legislation will significantly impact the future of AI in the state and beyond. The changes undoubtedly make the bill more palatable to AI firms, reducing potential trigger points for legal action. Nevertheless, the continued resistance indicates there is much to discuss regarding AI and the regulations we impose on it, thus signaling an ongoing, pivotal dialogue about our technological future.

While the passing of amended SB 1047 might potentially lessen the immediate liability of AI firms, the future of AI regulation in the United States — and by extension, the world — is far from settled.