P2P

16 "show your work" requirement that sheds light on an industry often criticized for its lack of transparency. Essentially, the law mandates disclosure of the careful considerations developers make as these models continue to evolve. The law also requires formal risk assessments for catastrophic harms, forcing companies to think seriously – and document explicitly – how their models might be misused or behave in dangerous ways. These assessments address worst- case scenarios: sophisticated cyberattacks, advanced biological threat generation, or unexpected autonomous behaviors. Some details remain internal, but the existence of these assessments and high-level summaries offers the public a window into the safety culture of frontier AI developers. Finally, if something goes wrong – if a model behaves dangerously, causes or contributes to a significant incident – SB 53 requires companies to report the issue to California's Office of Emergency Services. It is the AI- era equivalent of hazard-reporting frameworks in industries such as aviation, biomedicine, and energy. By establishing a feedback loop between developers and regulators, the law aims to identify patterns, emergent risks, or recurring vulnerabilities before they spiral into large-scale harm. One of SB 53's notable, and perhaps underappreciated, features is its inclusion of robust whistleblower protections. People building AI systems are often the first to spot dangerous shortcuts, underfunded safety processes, or unresolved hazards hidden in the architecture of a robust model. In a competitive, high-stakes field, internal pressure to "ship the model now and patch the safety later" can be immense. The result is a more accountable ecosystem – one where internal transparency supports public safety, and where workers feel empowered to speak out before small risks become global problems. CALCOMPUTE: AI INFRASTRUCTURE FOR THE MANY, NOT THE FEW Even more exciting is the introduction of CalCompute – essentially a frontier model for all! Today, compute access is the most significant barrier to AI innovation. Running a frontier-class model often requires resources that only big tech or well-funded labs can afford. CalCompute flips that dynamic. By providing startups, universities, and public institutions with access to powerful computing resources in a regulated, safe environment, California aims to broaden the AI ecosystem. It is a "community supercomputer," built on the idea that transformative technology should not be limited to billion-dollar corporations. FROM SILICON VALLEY TO THE GLOBAL STAGE The economic implications of SB 53 are vast, extending well beyond California's borders. Given that many of the world's leading AI firms are headquartered in the state, compliance with SB 53 could quickly become a de facto national standard. Rather than maintaining separate systems for California and the rest of the country, most companies will likely adopt SB 53's policies across their entire operations, effectively nationalizing its MORE ONLINE Did you miss the Fall issue? Read it online: https://epubs.iltanet. org/i/1540097 ó

• Cover