Anthropic Backs SB 53: A Blueprint for Balanced AI Safety and Innovation

Anthropic, the developer of AI model Claude, has endorsed California’s SB 53, a legislative effort aimed at governing the development and deployment of powerful artificial intelligence systems. The bill, introduced by State Senator Scott Wiener, mandates that large AI companies publish safety frameworks outlining how they assess and mitigate catastrophic risks, such as those that could lead to mass casualties or significant financial harm. Anthropic emphasized that the bill formalizes existing practices the company already follows, including the publication of Responsible Scaling Policies and comprehensive model system cards [1].

Governor Gavin Newsom had previously convened the Joint California Policy Working Group, composed of industry and academic experts, to advise on AI governance. The group’s recommendation of a “trust but verify” model is reflected in SB 53, which emphasizes transparency through disclosure requirements rather than prescriptive technical mandates. This approach contrasts with last year’s failed attempt, SB 1047, which faced criticism for imposing overly burdensome and unclear technical standards [1].

Under SB 53, covered companies must publish transparency reports detailing their risk assessments and mitigation efforts before deploying new AI models. They must also report critical safety incidents within 15 days and provide whistleblower protections for employees who expose potential safety failures or risks to public health. The bill includes exemptions for smaller companies with revenues under $500 million, a move Anthropic supported in order to avoid stifling innovation while still holding major developers to high safety standards [2].

The bill’s focus on transparency, according to Anthropic CEO Dario Amodei, aligns with the company’s long-held view that AI companies should be open about their safety measures. Anthropic has previously advocated for a federal regulatory framework but acknowledges that given the current political landscape, state-level legislation may be necessary to prevent a patchwork of regulations. Amodei noted that SB 53 strikes a balance between safety and innovation, and if passed, it could serve as a model for future federal legislation [1].

While Anthropic commended the bill’s direction, it also outlined areas for potential improvement. For example, the current threshold for covered AI models—based on computational power (FLOPS) used in training—may exclude some powerful systems. Anthropic has proposed refining definitions and providing more flexibility in how safety protocols are documented and shared. The company also emphasized the need for regulatory frameworks to evolve alongside AI technology, ensuring that rules remain relevant as models become increasingly advanced [1].

The passage of SB 53 is seen as a test of Newsom’s ability to navigate politically sensitive technology policy and maintain the support of major tech firms. The bill’s more targeted approach and reduced industry pushback compared to its predecessor have increased its chances of becoming law. If successful, it could serve as a blueprint for other states and contribute to a broader national conversation on AI governance [2].

Source:

[1] Anthropic is endorsing SB 53 (https://www.anthropic.com/news/anthropic-is-endorsing-sb-53)

[2] Anthropic backs California bill that could test Newsom on AI (https://www.politicopro.com)