← Back to feed

Illinois passed Senate Bill 315 through its House 110-0 on May 28, following a 52-5 Senate vote. Governor JB Pritzker posted support immediately after the House vote, signaling a near-certain signature. The bill takes effect January 1, 2027.

SB 315 is the first US law to require annual independent third-party safety audits specifically for frontier AI developers. It goes further than California’s SB 53 (which applies to training runs above 10^26 FLOPs) and New York’s RAISE Act (which mandates 72-hour incident windows and $1M penalties) by adding the external audit requirement as a standing, annual obligation.

What the Bill Requires

Safety frameworks: Frontier AI developers must publish and annually update plans detailing how they address severe or catastrophic risks from their models.

Third-party audits: Annual independent audits of safety practices — the first such mandate at the state level in the US.

Incident reporting: 72 hours for standard incidents; 24 hours when an incident poses imminent risk of death or physical harm.

Whistleblower protections: Employees who report safety concerns are explicitly protected.

Pre-deployment reports: Required documentation of model capabilities, intended use, and risk disclosures before major releases.

The bill defines “catastrophic risk” as models capable of mass harm or creating damages exceeding $1 billion through cyberattacks or malfunction beyond human control.

Industry Response

OpenAI and Anthropic both endorsed the bill — an unusual alignment with a regulator that most industry groups had quietly lobbied against. OpenAI spokesperson Jamie Radice called it “a thoughtful approach to issues like transparency, audits, and incident reporting.” Anthropic’s head of US state and local government relations, Cesar Fernandez, said it sets “enforceable accountability” as models grow more powerful.

The endorsement from both leading frontier labs is strategically legible: a third-party audit requirement that applies equally to all competitors eliminates the reputational asymmetry between labs that publish safety reports voluntarily and those that don’t. It also positions both companies favorably in the federal preemption debate — if state law is the floor, lobbying for weaker federal standards becomes harder to defend publicly.

The State Patchwork Is Now Three Layers Deep

Illinois joins a growing stack of state-level AI governance:

• California SB 53: In force since January 2026. Applies to companies with $500M+ revenue training on 10^26+ FLOPs. 15-day incident reporting, annual audits, $1M per violation.
• New York RAISE Act: 72-hour incident window, $1M penalties, effective January 2027.
• Illinois SB 315: Annual third-party audits, 72-hour (24-hour for imminent harm) incident window, effective January 2027.
• Connecticut: Governor Lamont signed two AI bills the same week — one for data broker deletion (modeled on California’s system), one for frontier model oversight with thresholds similar to Illinois.

Federal action remains stalled. The Trump administration’s voluntary framework dropped mandatory pre-release testing requirements in April, leaving states as the de facto regulators. With four jurisdictions now passing meaningful AI governance, and Illinois’ 110-0 House vote signaling bipartisan momentum, the pattern has enough critical mass to drive compliance infrastructure regardless of what Congress does.

The practical implication for frontier labs: annual third-party audits are coming. The question is whether those standards end up defined by Illinois, California, New York, or a federal framework built from the state precedents. The labs that endorsed SB 315 have already decided they’d rather help write the standard than fight it.