Safe harbor
Affirmative compliance defenses under specified conditions.
How the laws define this
Safe harbor provisions across these instruments fundamentally agree on providing covered entities and developers protection from liability, penalties, or civil enforcement when specific qualifying conditions are satisfied. Where these definitions diverge is in the precise factual triggers required to earn that shield. Under Texas HB 149 — Texas Responsible Artificial Intelligence Governance Act (TRAIGA), affirmative defenses depend on substantial compliance with recognized frameworks like the NIST AI Risk Management Framework, finding violations through testing or good-faith audits, or experiencing third-party misuse. In contrast, Utah SB 226 - Artificial Intelligence Consumer Protection Amendments grounds its liability protections on making clear and conspicuous disclosures regarding generative artificial intelligence at the outset of and during consumer interactions. Meanwhile, the Take Responsibility for Unlocking the Machine Potential of America’s AI Act specifically shields developers from civil liability when third parties use their models for unlawful acts, provided there was no intentional design for such unlawful use.
Synthesised from the 3 statutory definitions below. Each one is quoted in full, with its source.
How jurisdictions differ
What is distinctive in each definition, binding law in force first.
| Jurisdiction | Instrument | Status | What is distinctive |
|---|---|---|---|
| Texas, United States | Texas HB 149 — Texas Responsible Artificial Intelligence Governance Act (TRAIGA) | In Force | Expands affirmative defenses to three specific circumstances: third-party misuse, good-faith audits, and substantial compliance with NIST AI RMF or equivalents. |
| Utah, United States | Utah SB 226 - Artificial Intelligence Consumer Protection Amendments | In Force | Limits protection to entities making clear and conspicuous disclosures about generative AI in consumer interactions. |
| United States | TRUMP AMERICA AI Act | Draft | Shields developers from civil liability for unlawful uses by others, if not intentionally designed for such acts. |
What it triggers
- In Texas under TRAIGA, it provides affirmative defenses against liability for prohibited AI uses.source
- In Utah under SB 226, it offers protection from liability or penalties in consumer interactions.source
- Under the TRUMP AMERICA AI Act draft, it would shield developers from civil liability for unlawful uses by others.source
Definitions (3)
A set of affirmative defenses available to entities in three circumstances: when prohibited uses result from third-party misuse of an AI system; when violations are discovered through testing or good-faith audits; and when entities substantially comply with the NIST AI Risk Management Framework or similar recognized standards.
A provision that offers protection from liability or penalties for entities that make clear and conspicuous disclosures about the use of generative artificial intelligence at the outset of and during consumer interactions.
Provisions that shield AI developers from civil liability if their models are used by others for unlawful acts, provided there was no intentional design for such use.
Not to be confused with
Safe harbour provides limited administrative leniency for supervised testing, unlike safe harbor's broader liability defenses.
Questions and answers
What defenses does safe harbor provide under Texas TRAIGA?
It offers affirmative defenses when prohibited uses result from third-party misuse, violations are found through testing or audits, or entities substantially comply with NIST AI RMF or similar standards. This applies to AI developers and deployers.
How does Utah's safe harbor work for generative AI?
Entities receive protection from liability or penalties by making clear and conspicuous disclosures about generative AI use at the outset and during consumer interactions.
What liability protection does the TRUMP AMERICA AI Act draft offer?
It shields AI developers from civil liability if their models are used by others for unlawful acts, provided there was no intentional design for such use.
Related Terms
Safe harbour
Limited administrative leniency for supervised testing....
Covered AI
AI developed, used, or acquired by or on behalf of agencies....
Safe AI
AI designed to prevent unintended harm across its lifecycle....
Low‑Risk AI
AI systems subject to baseline transparency and documentation....
High-risk artificial intelligence system
An AI system that substantially factors in consequential decisions....