AI Frontier Institute

AI MODEL SAFETY

View primary source →

Illinois SB3444, the Artificial Intelligence Safety Act (Section 1), creates a liability safe harbor for developers of frontier AI models. Under Section 10, a developer is not liable for 'critical harms' — defined in Section 5 as the death or serious injury of 100 or more people or at least $1 billion in property damage caused by CBRN weapons or autonomous criminal conduct — if the developer did not intentionally or recklessly cause those harms and publishes both a safety and security protocol (per Section 15) and a transparency report (per Section 20) on its website prior to or at model release. A 'frontier model' is defined in Section 5 as one trained using more than 10^26 computational operations or costing more than $100 million to compute. Under Section 10(b), developers can satisfy these requirements by either agreeing to be bound by Article 56 of the EU AI Act or by entering a federal agency agreement enabling model access, evaluation of cyber and biological risks, and public release of evaluation findings. Developers choosing the federal agreement path must file a certification with the Illinois Attorney General under Section 10(c). Section 15 details the safety and security protocol requirements, including testing procedures, risk thresholds, mitigation strategies, third-party assessments, cybersecurity practices, and monitoring plans; redactions to protect trade secrets are permitted. Section 20 requires transparency reports identifying the model and summarizing assessment results and risk-mitigation steps, also allowing redactions. Section 25 provides that the Act sunsets if the federal government enacts overlapping requirements for frontier model developers, and deems developers compliant if they comply with a substantially similar framework in another state.

← Back to the tracker