State legislators nationwide have proposed or passed a flurry of laws to regulate AI, with new ideas continuing to emerge as most state legislatures prepare to convene.
The most notable bill to date is the Texas Responsible AI Governance Act (TRAIGA). It would be the most expansive state regulation of AI if the current version becomes law.
TRAIGA would take a risk-based approach to governance modeled after the EU AI Act, creating obligations for developers, deployers and distributors of systems deemed high-risk. It proposes a more expansive definition of “high-risk,” a ban on certain AI systems and a narrow private right of action.
Here are five things to know about the bill:
1. It would create obligations for developers and deployers of any AI system that is a “contributing factor” for high-risk decisions.
- Decisions relating to employment, finance, education, health-care services, housing and insurance could constitute “high risk” decisions.
- Texas would be the first state to also require high-risk AI system distributors to take “reasonable care” to prevent algorithmic discrimination.
2. The bill would ban AI systems that pose “unacceptable risk.”
- Much like the EU AI Act, TRAIGA would restrict certain uses of AI.
- It would prohibit systems that – without express consent – identify emotions, capture biometric identifiers or use such identifiers to categorize consumers based on sensitive attributes.
3. Private litigants would have a limited right of action.
- The attorney general would enforce most provisions of TRAIGA, but private litigants could go to court over alleged violations that implicate banned AI systems.
4. The law would require generative AI developers to “keep detailed records” of datasets used to train their models.
5. The law would exempt many small businesses and provide limited carveouts for innovation and experimentation.
- Companies that meet the Small Business Administration definition of “small business” would have no obligations under TRAIGA.
- Developers of novel AI could apply for similar exemptions under a time-limited experimental sandbox program.
- Companies investigated under TRAIGA would have 30 days to cure alleged violations before the attorney general can bring an action.
What’s Next?
Legislators in Colorado favored a risk-based governance framework in passing the Colorado AI Act this year, just as would the Texas bill.
Yet the Texas bill would go beyond what Colorado did by proposing a more expansive definition of “high risk,” banning “high risk” AI systems and providing a narrow private right of action.
Only time will tell if the Texas bill survives and advances in its current form, but AI oversight bills often narrow in scope as they go through the legislative process. This was true of Colorado’s AI Act and will likely hold true for TRAIGA, as well.