AN ACT CONCERNING ARTIFICIAL INTELLIGENCE.
What changed between versions
The definition of 'consequential decision' was modified to emphasize 'access to' employment, education, and government services, clarifying that the bill applies to decisions affecting access rather than just decisions made about those areas.
The definition of 'high-risk artificial intelligence system' was significantly expanded with seven specific exclusions, including anti-fraud technology without facial recognition, video game technology, cybersecurity tools, internal management systems, document classification systems, decision pattern detection tools, and natural language communication tools.
The definition of 'deploy' was changed from making consequential decisions to simply putting a high-risk AI system into use, broadening the scope of what constitutes deployment.
The definition of 'Integrator' was updated to clarify that integrators neither develop nor substantially modify the AI system but integrate it into products or services offered to others.
The definition of 'General-purpose artificial intelligence model' was restructured to specify criteria for generality, task range, and downstream application integration, while excluding models used only for development, prototyping, and research before market release.