SB 3444 — Artificial Intelligence Safety Act (Frontier Liability Framework)

Frontier AI developers (any model trained with $100M+ in compute) would have to publish a safety framework and report 'critical harms' — but would NOT be liable for mass-casualty events of 100+ deaths, $1B+ property damage, or CBRN-weapon misuse caused by their models. OpenAI is backing it; Anthropic is opposing.

What it does

Establishes a 'critical harm' liability framework for frontier AI developers: defines critical harms as events causing death/serious injury to 100+ people, $1B+ in property damage, or use of AI to develop chemical, biological, radiological or nuclear weapons. Coverage tied to a $100M training-compute threshold. Mandates published safety frameworks and incident reporting, but creates a liability shield against mass-casualty claims for compliant developers.

Analyst read

UPDATE (June 2026): Cunningham’s OpenAI-aligned liability framework did not advance. The legislature instead passed Sen. Edly-Allen’s framework through the SB 315 vehicle, which omits SB 3444’s controversial mass-casualty liability shield. After the public OpenAI/Anthropic split drew national attention, the negotiated SB 315 dropped the private right of action and extended operational deadlines, and both labs ultimately stood down. Original analysis: the liability shield was the politically toxic provision; it was stripped rather than folded into the enacted-track bill.

Enforcement

Civil penalties for safety-framework violations; liability shield for compliant developers

Procedural history

2026-06-01 Did not advance; spring session adjourned. Frontier-safety legislation passed via SB 315 instead The catastrophe liability-shield framework was not adopted

2026-04-24 Rule 2-10 Committee/3rd Reading Deadline Established As May 15, 2026 Second extension; survives without committee vote

2026-04-17 Fortune publishes 'Illinois is OpenAI and Anthropic's latest battleground' National media attention may slow the bill as legislators reassess

2026-04-15 Anthropic publicly opposes the bill; lobbies Sen. Cunningham to revise or reject Frontier-lab split signals the liability-shield language is the wedge issue

2026-04-09 OpenAI testifies in support at Senate AI subject-matter hearing First-ever direct testimony by an AI lab in front of an Illinois committee

2026-03-27 Rule 2-10 Committee Deadline Established As April 24, 2026 First extension

2026-03-13 Rule 2-10 Committee Deadline Established As March 27, 2026

2026-02-18 To AI and Social Media

2026-02-17 Assigned to Executive

2026-02-04 Filed by Sen. Bill Cunningham; First Reading; Referred to Assignments Same filing day as Edly-Allen's safety bills — competing Senate frameworks

Artificial Intelligence Safety Act (Frontier Liability Framework)

What it does

Analyst read

Enforcement

Procedural history

Regulatory vectors

Primary sources 1