California has softened a bill aimed at preventing AI-related disasters following recommendations from Anthropic, as it approaches the final vote.

California's bill SB 1047, designed to prevent AI-related disasters, has encountered strong opposition from various Silicon Valley stakeholders. In response, California lawmakers have made several amendments, influenced by AI firm Anthropic and other critics. On Thursday, the bill passed through the state's Appropriations Committee, a significant step towards becoming law, with key changes incorporated, according to Senator Wiener’s office in a statement to TechCrunch.

Senator Wiener expressed that they accepted several reasonable amendments proposed by industry members, addressing core concerns raised by Anthropic and others. He also mentioned that earlier modifications were made to consider the needs of the open-source community, a critical innovation source.

Despite the amendments, SB 1047 still seeks to hold developers liable for large-scale AI systems that could potentially cause mass casualties or cybersecurity incidents costing over $500 million. However, the revised bill now reduces the extent of California's government power to hold AI laboratories accountable.