Bill > SB3444

This bill, titled the Artificial Intelligence Safety Act, aims to shield developers of "frontier artificial intelligence models" – which are extremely powerful AI systems trained with vast computational power or at a high cost – from liability for "critical harms" (defined as significant loss of life or property damage caused by the AI, particularly through weapons or actions without human intervention that would be criminal if done by a person) if they did not intentionally or recklessly cause these harms. To qualify for this protection, developers must publish a detailed safety and security protocol outlining how they manage, assess, and mitigate risks, along with a transparency report detailing the results of their risk assessments and the steps taken to address them, both on their website. Developers can also meet these requirements by agreeing to abide by the European Union's AI safety standards or by entering into an agreement with a federal agency that allows for government access to their models for research and evaluation. The bill also specifies that it will cease to apply if the federal government enacts its own overlapping regulations for frontier model developers.