California Governor Gavin Newsom has signed SB 53, the first U.S. law requiring leading AI companies such as OpenAI, Anthropic, and Meta to publicly disclose their safety measures and report potential risks tied to the development of advanced AI systems.
Under the new legislation, companies must report any incidents involving deceptive AI behavior or other major safety threats to the California Office of Emergency Services within 15 days. The law also includes whistleblower protections, allowing employees to report potential risks without fear of retaliation.
The regulations apply only to firms with annual revenues exceeding $500 million and to AI models surpassing 10²⁶ FLOPs in processing power — exempting smaller startups. While Anthropic supported the bill, OpenAI and Meta initially lobbied against it, warning of complications if each U.S. state enacts separate rules. However, both companies ultimately agreed to comply.
Given that 32 of the world’s 50 largest AI companies are based in California, SB 53 has the potential to become a global precedent. Unlike the previously rejected SB 1047, which demanded an AI “kill switch,” the new law focuses on transparency rather than restriction — offering a balanced framework that seeks to merge innovation with accountability.
In brief: Tech World Highlights:
- Exelixis, a California-based biotech company specializing in anti-cancer therapies, is laying off 130 employees, consolidating operations, and closing its Pennsylvania facility.
- Novo Nordisk has secured global rights to Replicate Bioscience’s self-replicating RNA technology in a $550 million deal, targeting new treatments for obesity and diabetes.
- Finland has launched the world’s largest industrial sand battery in the city of Pornainen, replacing a wood-chip plant and aiming to cut local heating emissions by up to 70%.
- California has reached a historic agreement with Uber and Lyft, allowing drivers to unionize while maintaining their independent contractor status.
- Anthropic now requires all Claude users to choose by September 28 whether they consent to their conversations being used for model training.
AI Trending Tools:
- Co-STORM – AI tool that generates Wikipedia-style articles from scratch with integrated search capabilities.
- Hunyuan-A13B – Tencent’s new open-source model for hybrid reasoning.
- Qwen VLo – Alibaba’s GPT-4o-like model for image generation and editing.
