Illinois Proposes New Rules to Make AI Companies Safer

Illinois lawmakers have introduced new laws aimed at making artificial intelligence safer and clearer about who is responsible when things go wrong. The package targets companies building the largest and most powerful AI systems — the kind of AI that companies like OpenAI and Google are working on. One state lawmaker, Senator Mary Edly-Allen, is leading the effort with Senate Bill 3312.

The Main Law: SB3312

The central proposal, Senate Bill 3312, creates what's called the Artificial Intelligence Safety Measures Act. The law focuses on the biggest developers of cutting-edge AI — what the bill calls "large frontier developers" — and requires them to write down and share their safety plans publicly.

These safety plans would need to address three main things: how they manage the risk of really bad outcomes, how transparent they are about how their AI works, and how they protect their AI systems from being hacked or stolen.

This approach is narrower than a similar law California tried to pass, which faced pushback from tech companies worried about the cost and difficulty of following such rules. Illinois is betting that focusing only on the biggest developers will be easier to implement.

Other Related Bills

The package includes a few other bills working alongside SB3312. One bill, Senate Bill 3384, creates another set of safety rules, though the details are still being worked out in committee.

A third bill, Senate Bill 3444, does something different: it shields AI developers from lawsuits if their AI system causes serious harm. This is unusual because it puts safety requirements in one law while protecting companies from legal blame in another. The idea appears to be that because AI systems can behave in unexpected ways, it is hard to know who to blame when something goes wrong.

Protecting People Who Speak Up

Senate Bill 315 adds a new protection to Illinois's whistleblower law. It protects workers who report safety problems at AI companies from being fired or punished for doing so. The law recognizes that people working inside these companies may spot safety failures that outside regulators would never see.

Building on Past AI Rules

Illinois has already passed laws about AI in one area: the workplace. In 2024, the state passed a law preventing companies from using biased AI when hiring. Representative Eric Sorensen also introduced a law dealing with deepfakes — fake videos of real people — in early 2024. So Illinois is taking a multi-part approach to different AI risks.

What Still Needs to Be Decided

The bills raise some practical questions that lawmakers and agencies will need to work out later. For example, what exactly counts as a "large frontier developer." How powerful does an AI system have to be before these rules kick in. And what should these safety plans actually look like.

The cybersecurity part is worth paying attention to. If the law requires companies to protect their AI systems from being stolen or hacked — which it appears to — that could change how major AI labs are set up and run.

Why This Matters Beyond Illinois

The broader context here is that states have always moved faster than the federal government when a new technology comes along. Back in the 1990s, different states tried different approaches to regulating the internet before any federal rules existed. We are seeing the same thing now with AI. Illinois is one of the first states to try, and what works or doesn't work here could shape what other states do next.

The part of these bills that might matter most is the liability shield in SB3444. If it holds up in court, it could make Illinois an attractive place for AI companies to set up shop, because they would face less risk of being sued. At the same time, the safety requirements would impose real costs on companies trying to comply.

One practical question: these laws will only affect companies that do business in Illinois or are based there. The biggest AI companies operate everywhere, so they would likely have to follow Illinois rules anyway. But smaller AI startups might choose to locate elsewhere to avoid the burden. It remains to be seen whether these laws actually change how AI gets developed, or whether they end up being more symbolic than practical.