HB 3545: Technology; artificial intelligence; state agencies; prohibited uses; permitted uses; Office of Management and Enterprise Services; effective date. - Oklahoma

House Committee Substitute → Floor (House) · 3 edits

MINOR

The bill was updated from a House Committee Substitute version to the Floor version, incorporating additional sponsors and refining the formatting of definitions and prohibited uses. The substantive policy content regarding AI regulations remains consistent, with no changes to the core legal requirements or scope.

Scope change

The bill's scope and applicability remain unchanged; the text represents a procedural update to the legislative record rather than a policy modification.

TECHNICAL

Added names of two additional House sponsors (Hill and Stewart) to the list of bill authors.

Adjusted page headers and footers to reflect the 'Floor Version' status and added a note indicating that bold face text denotes committee amendments.

Reformatted the layout of definitions and prohibited uses to align with standard floor version formatting, removing internal committee markup indicators.

Proposed Policy Committee Substitute 1 → House Committee Substitute · 6 edits

MODERATE

The bill was completely rewritten from scratch, replacing the original text with a new version that updates terminology and clarifies legal definitions. This rewrite ensures the legislation aligns with specific committee amendments, particularly refining how 'Artificial Intelligence' and 'Deepfakes' are defined and adjusting the exceptions for law enforcement use of biometric tools.

Scope change

The scope of the bill remains focused on regulating AI use by state agencies, but the applicability of certain definitions and the specific wording of exemptions for law enforcement have been modified to be more precise.

DEFINITION

The definition of 'Artificial intelligence' was changed from a general description of capabilities to a specific definition of an 'artificial intelligence system' that infers outputs from input, explicitly excluding robotic process automation.

The definition of 'Deepfake' was updated to clarify that it includes videos where a person appears to say or do things they never actually did.

The definition of 'High-risk AI decision' was refined to specify that it includes automated determinations affecting legal rights or benefits that lack meaningful human review before implementation.

A new definition for 'Generative artificial intelligence system' was added to specifically cover systems designed to create synthetic content like text, images, or video in response to prompts.

REQUIREMENT

The requirement for biometric identification systems was updated to allow use in public spaces only when there is a valid warrant, court order, or specific exigent circumstances like an active shooter or credible terrorist threat.

TECHNICAL

The bill title and header were updated to reflect the new committee substitute version, including a revised list of sponsors and a new amendment reference number.

Proposed Committee Substitute (full committee) 1 → Proposed Policy Committee Substitute 1 · 4 edits

MODERATE

The bill was renamed from an 'Oversight' bill to a 'Policy' bill, indicating a shift in focus from monitoring to setting rules. The most significant change is the redefinition of 'Artificial Intelligence,' which now emphasizes human-like cognitive capabilities (reasoning, learning) rather than just output generation. Additionally, the requirements for biometric identification were tightened, restricting their use to locating missing persons or specific emergency threats, while the definition of 'Generative AI' was simplified to focus on generating text, images, or media.

Scope change

The bill's scope shifted from a general oversight framework to a specific policy framework with stricter definitions of AI capabilities and more limited exceptions for biometric surveillance.

DEFINITION

The definition of 'Artificial Intelligence' was changed from systems that generate outputs to systems that display human-like cognitive capabilities such as reasoning, learning, planning, and creativity.

The definition of 'Generative AI' was simplified to mean AI that generates text, images, or other media in response to prompts, removing the exclusion for basic editing functions.

REQUIREMENT

The exception for real-time biometric identification (like facial recognition) was narrowed significantly. It is now only allowed to locate missing or abducted persons or during specific emergency threats (e.g., active shooters), removing the previous exception for valid warrants or court orders.

SCOPE

The bill title and header were updated from 'Proposed Oversight Committee Substitute' to 'Proposed Policy Committee Substitute', reflecting a change in the bill's legislative intent.

Engrossed → Proposed Committee Substitute (full committee) 1 · 3 edits

MINOR

The bill was reformatted from an 'Engrossed' version to a 'Proposed Committee Substitute' version, primarily updating the header to reflect the committee's specific amendments and changing the legislative session date to 2026. The substantive policy language regarding AI definitions, prohibited uses, and reporting requirements remains identical to the previous version.

Scope change

The scope of the bill's applicability and the specific policy requirements for state agencies have not changed; only the document metadata and formatting have been updated.

TECHNICAL

The document header was changed from 'Engrossed House' to 'Proposed Oversight Committee Substitute' to indicate the bill is in a specific committee review stage.

The legislative session date was updated from an unspecified date to the 2nd Session of the 60th Legislature (2026).

The text was renumbered and reformatted to align with standard committee amendment markup conventions.

Introduced → Engrossed · 6 edits · Mar 11, 2026

MODERATE

The bill was amended to modernize its language by replacing the term 'artificial intelligence' with 'artificial intelligence system' to ensure legal precision. Additionally, the definitions for 'deepfakes' and 'high-risk AI decisions' were expanded to include specific criteria regarding malicious use, legal impacts, and the necessity of human review before implementation.

Scope change

The scope of prohibited uses was narrowed and clarified; for instance, facial recognition is now explicitly allowed for locating missing persons and under valid warrants, whereas previous versions were broadly prohibited except for missing persons.

DEFINITION

The definition of 'Artificial intelligence' was changed to 'Artificial intelligence system' to define it as a machine-based system that infers outputs from inputs, rather than just the ability to display capabilities.

The definition of 'Deepfake' was updated to use gender-neutral language ('his or her') and clarified that it includes actions the person appears to be doing that they have never done.

A new definition for 'High-risk AI decision' was added, defining it as automated determinations affecting legal rights, benefits, or civil penalties that require meaningful human review.

REQUIREMENT

Prohibitions on facial recognition were modified to allow use when conducted pursuant to a valid warrant, court order, or in exigent circumstances involving an imminent threat to life.

The requirement for human review was clarified to mandate that any high-risk AI decision must be reviewed and approved by a human employee with decision-making authority before taking effect.

The requirement for disclosure was updated to specify that materials generated by AI must be accompanied by a disclosure stating the content was generated by AI.

Technology; artificial intelligence; state agencies; prohibited uses; permitted uses; Office of Management and Enterprise Services; effective date.

What changed between versions

How they voted

Actions timeline

Sponsors