What happened
The Illinois legislature passed SB 315 this week, a significant AI safety law. According to Ars Technica, the bill requires large AI firms to submit public safety plans and annual reports from independent third-party auditors. Companies must also report critical safety incidents to the state within 72 hours.
Governor J.B. Pritzker has confirmed he will sign the bill into law. Its provisions are set to take effect on 1 January 2027.
How the room's reading it
The bill's passage has split opinion. Frontier labs like OpenAI and Anthropic publicly supported the legislation, framing it as a way to standardise safety practices they already follow and avoid a messy patchwork of state-level rules. Some policy watchers, however, suggest this move benefits incumbents who can easily absorb compliance costs, potentially creating barriers for smaller AI firms. Trade groups like the Chamber of Progress have been more critical — raising concerns about forcing companies to expose sensitive systems to untested auditors. Meanwhile, the bill's sponsors in Illinois position the law as a necessary stopgap while Congress stalls.
Sailfish's take
We're less interested in the politics and more in the operational reality this creates for builders. The Illinois law itself is manageable — it's the precedent that matters. We're now entering an era of state-by-state AI compliance, which means legal and operational overhead is about to become a significant cost centre for anyone shipping models. We've seen this play out with privacy laws. The real moat for frontier models might not be parameter count or training data, but the size of your policy team. For now, we'd advise any team working on foundation models to start tracking state legislatures. This isn't just a big-lab problem anymore.