What happened
OpenAI initially supported Illinois Senate Bill 3444, which proposed shielding frontier AI developers from civil liability for catastrophic harms exceeding 100 deaths or $1 billion in property damage, provided safety protocols were published and harm was not intentional or reckless. The company now explicitly opposes this safe harbour, instead backing Senate Bill 315, a transparency framework that removes the liability shield entirely. Anthropic consistently opposed the original shield. SB3444 applied to models trained with over 10^26 FLOPs or $100 million compute cost, covering major developers.
Why it matters
Direct civil liability now applies to frontier model developers for catastrophic harms, even with published safety protocols, increasing risk profiles for model deployment. Procurement teams and legal architects must scrutinise model capabilities and audit reports for potential liabilities. This contrasts with earlier legislative proposals, such as Colorado's framework, which explored different liability shifts, and mandates closer attention to explicit civil liability mechanisms in state legislation.
