California Senate Bill 53 proposes a set of safety and transparency requirements for companies developing advanced artificial intelligence systems. The bill focuses on organizations building large-scale models that require significant computing power and have the potential to cause widespread harm. These risks include cyberattacks, misuse in biological research, and other incidents that could result in mass casualties or financial damage.
The bill outlines several obligations for qualifying companies. These include publishing safety frameworks, submitting risk assessments, and reporting critical incidents to state authorities. Developers must also provide whistleblower protections and maintain anonymous reporting channels for internal concerns.
Anthropic, a San Francisco-based AI company, has publicly endorsed the bill. The company stated that the proposed regulations align with practices it already follows, such as publishing system cards and maintaining a Responsible Scaling Policy. These documents describe how the company tests its models and mitigates risks associated with powerful AI systems.
The endorsement comes after a previous version of the bill was vetoed. In response, California lawmakers revised the proposal to focus more narrowly on transparency rather than liability. The updated version reflects recommendations from a working group convened by the governor, which included academic and industry experts.
Anthropic’s support for SB 53 contrasts with opposition from other tech firms and trade groups. Organizations such as the Consumer Technology Association and the Chamber of Progress have expressed concern that the bill could discourage innovation or push companies to relocate. Some investors have argued that regulation should be handled at the federal level to avoid conflicting state laws.
Despite these objections, the bill has gained traction in the California Legislature. It has passed several voting rounds and is scheduled for a final decision. If approved, it would apply only to companies with annual revenues above $500 million and models trained with extensive computing resources.
The bill does not impose requirements on smaller startups or companies developing less powerful systems. This distinction is intended to avoid placing unnecessary burdens on early-stage developers while focusing oversight on the most capable and potentially risky models.
Anthropic has acknowledged that federal regulation would be preferable but described California’s action as necessary given the pace of AI development. The company emphasized that powerful models are already being built and deployed, and that waiting for national consensus could leave gaps in oversight.
Safety Measures and Reporting Requirements
SB 53 includes several specific measures aimed at improving public safety. Companies must publish safety frameworks that explain how they assess and manage risks. These frameworks should describe testing procedures, thresholds for dangerous capabilities, and mitigation strategies.
Before releasing a new model, developers must submit a public transparency report. This document should summarize the results of risk assessments, explain deployment decisions, and disclose whether third parties were involved in testing. In cases of critical incidents, such as loss of model control or data leaks, companies must notify the state within 15 days. Urgent cases require notification within 24 hours.
The bill also mandates confidential reporting of catastrophic risk assessments to California’s Office of Emergency Services. These assessments evaluate the potential for large-scale harm and are intended to help the state prepare for possible emergencies.
Whistleblower protections are a key component of the legislation. Employees who report safety concerns must be shielded from retaliation, and companies must provide secure channels for anonymous disclosures. These provisions aim to encourage transparency and accountability within AI development teams.
If passed, SB 53 would establish California as the first state to enforce safety and transparency standards for frontier AI systems. The bill could influence how other states and federal agencies approach regulation, especially as AI technologies continue to expand into new sectors.
The legislation also reflects California’s broader role in shaping technology policy. With many leading AI companies headquartered in the Bay Area, the state has a direct stake in how these systems are developed and deployed. The bill’s focus on transparency rather than prescriptive technical mandates may offer a flexible model for oversight without stifling innovation.
Anthropic’s endorsement suggests that some developers are willing to accept regulatory obligations in exchange for public trust. By formalizing voluntary practices, the bill aims to create a consistent baseline for safety across the industry.
While debate continues among lawmakers, investors, and developers, the bill’s progress signals a shift toward more structured oversight of advanced technologies. For California residents and businesses, the outcome may shape how AI systems are integrated into daily life, public services, and economic planning.
The final vote on SB 53 is expected soon. If approved, the bill will become a reference point for future discussions about AI governance, both within California and beyond. For now, the endorsement from a major San Francisco-based developer adds weight to the proposal and may influence its path forward.
