A real estate broker or salesperson, or person acting on their behalf, who includes a digitally altered image in any advertisement or promotional material for the sale of real property must include a reasonably conspicuous disclosure statement on or adjacent to the image stating that the image has been altered. The disclosure must also include a link to a publicly accessible internet website, URL, or QR code where the original, unaltered image is clearly identified and available. The statement must include language indicating that the unaltered images can be accessed at the linked location.

When an advertisement or promotional material containing a digitally altered image is posted on an internet website controlled by the real estate broker or salesperson (or person acting on their behalf), the unaltered version of each digitally altered image must also be included in the posting. Compliance may be achieved by including a link to a publicly accessible internet website that includes and clearly identifies the original, unaltered image. If the link method is used, the disclosure statement required by subdivision (a)(1) must include language indicating the unaltered images can be accessed on the linked website, URL, or QR code.

Persons distributing synthetic media depicting a candidate or party within 90 days of an election must include a disclosure stating the content has been manipulated or generated by artificial intelligence — in easily readable text (at least as large as the largest other text) for the duration of any video, or clearly spoken at the beginning, end, and at two-minute intervals for audio-only media.

Any person must include a clear and conspicuous disclosure identifying content as AI-generated before distributing within Tennessee (1) an image, video, or audio recording using another individual's likeness, or (2) literature or other written work attributed to someone other than the actual author — regardless of whether consent was obtained and regardless of whether the content has been modified after generation.

Any person must include a clear and conspicuous AI-generated disclosure, appropriate for the medium, on AI-generated content that uses another individual's likeness (image, video, audio, or song) or that constitutes a written work attributed to someone other than the actual author, prior to distributing that content in Tennessee — regardless of consent and regardless of post-generation modifications.

Publishers of synthetic media may assert an affirmative defense by including a prescribed disclosure statement — This (image/video/audio) has been manipulated with the use of artificial intelligence or another digital technology and is not an accurate representation of the (individual/event) depicted — displayed in easily readable text no smaller than the largest other font in visual media (for the full duration if video), or spoken clearly at the beginning, end, and at two-minute intervals for audio-only media.

Developers of high-risk generative AI systems must ensure that synthetic content outputs are identifiable and detectable using industry-standard tools or developer-provided tools, with identification applied at the time of generation. For artistic, creative, satirical, or fictional works, identification must not hinder display or enjoyment. Text-only outputs, public-interest informational content, content unlikely to mislead a reasonable person, and assistive editing outputs are exempt.

Suppliers must disclose the use of artificial intelligence technology in the creation of any videographic or still image intending to depict an actual person, or any audio or audio-visual recording intending to depict the voice of an actual person, when the supplier disseminates or sells the item.

Any person who publicly distributes a realistic digital imitation must include a clear and conspicuous disclosure that it is AI-generated or AI-altered — visible for images and video, audible for audio, and expressed in plain language.

Any person who produces or creates a commercial advertisement must conspicuously disclose in the advertisement that a synthetic performer appears in it, where the person has actual knowledge of the synthetic performer's use.

Any person who includes a digitally altered image (including AI-altered images) in an advertisement or promotional material for the rental of real property must include a reasonably conspicuous statement disclosing that the image has been altered. The statement must be located on or adjacent to the image and must include a link, URL, or QR code to a publicly accessible website that includes and clearly identifies the original, unaltered image.

When a digitally altered rental property advertisement is posted on an internet website over which the person has control, the person must also include the unaltered version of the images directly in the posting. The person may satisfy this requirement by including a link to a publicly accessible website that includes and clearly identifies the original, unaltered images. If compliance is via link, the disclosure statement required under subdivision (b)(1) must indicate that unaltered images can be accessed at the linked website, URL, or QR code.

Any person who creates and causes to be published in an advertising medium a video or audiovisual advertisement that includes a synthetic performer must include a clear and conspicuous disclosure that the performer is synthetic. The disclosure must: (1) appear in close proximity to the synthetic performer, (2) be presented in a manner and duration sufficient for a reasonable consumer to understand that the performer is not a real human being, and (3) use wording substantially similar to 'this performance features a synthetic digital performer' or 'no human performer is depicted.' Exempt from this requirement are advertisements for expressive works (motion pictures, television programs, streaming content, documentaries, video games, or similar audiovisual works) where the synthetic performer's use in the advertisement is consistent with its use in the expressive work, and advertisements where AI is used solely for language translation of a human performer.

Any news media content substantially created by generative AI that is published, broadcast, or otherwise disseminated or accessible in New York must carry a conspicuous disclosure that it was substantially created by generative AI. For visual and audiovisual content, the disclosure must be imprinted at the top of the page, webpage, image, graphic, or video. For audio content, the disclosure must be verbally stated at the onset. This requirement does not apply if the content is eligible for copyright registration.

Developers must program AI systems to provide a distinctive watermark on any AI-generated product that informs the user that the product was generated using an AI system.

No person may remove a watermark required by this section with the purpose of concealing that the product was generated using artificial intelligence.

Developers must include a clear and conspicuous disclosure identifying output from any generative AI system producing images, video, or audiovisual content as AI-generated content. The disclosure must include a human-perceptible label and embedded metadata identifying the content as AI-generated, the tool used, and the creation timestamp. The disclosure must, to the extent technically feasible, be permanent or unable to be easily removed. Visual-only content requires visual disclosure; audiovisual content requires both visual and audible disclosure. All disclosures must stand out by size, contrast, location, and duration; be unavoidable; use plain language; and not be contradicted by the content itself.

Any individual disseminating AI-generated content (video, audio, or still images) to facilitate commerce or trade must include a disclosure that artificial intelligence was used to make all or part of the content.

Any individual disseminating AI-generated visual media in commerce must ensure the disclosure is (1) written in each language used in the content and (2) clearly readable — appearing in letters at least 30% of vertical picture height, visible for at least 30% of the media's length for moving images, and rendered with adequate color contrast.

Any individual disseminating AI-generated audible media in commerce must ensure the disclosure is announced at the same volume, speaking rate, and in each spoken language as the content itself.

Any person operating an artificial intelligence system that produces images, videos, audio, or multimedia AI-generated content must include on that content a clear and conspicuous disclosure identifying the content as generated by artificial intelligence.

Generative AI systems used within Massachusetts must include on or within AI-generated content (1) a clear and conspicuous human-perceptible notice identifying the content as AI-generated, permanent or difficult to remove to the extent technically feasible, and (2) embedded metadata identifying the content as AI-generated, the identity of the generating system, and the date and time of creation.

Any person within Massachusetts who uses a generative AI system to create or repurpose AI-generated content must not remove any AI-disclosure notice or metadata required by Section 2.

Consumer reporting agencies must ensure that AI system outputs — including any synthetic audio, image, video, or text — are detectable as artificially generated.

Publishers and disseminators of AI-generated or AI-modified images, video, or audio may label such content with a visible or audible mark or statement developed by the Secretary of State indicating that the content was generated or modified using artificial intelligence. Failure to label triggers civil liability under subsection (3) if a person is injured as a result.

Publishers and disseminators who fail to place a mark or statement on AI-generated or AI-modified images, video, or audio indicating AI origin are subject to a private cause of action for damages by any person who suffers an injury as a result of the unlabeled content.

Any person or entity creating AI-generated content for public consumption must clearly label the content as AI-generated in a prominent and conspicuous manner, identifying the AI system used and, if applicable, the developer or deployer, with format-specific labels: header/footer for text, verbal disclosure at start and every two minutes for audio exceeding 30 seconds, and visible watermark or overlay with metadata and captions for images and video.

All labels and watermarks required under the act must be accessible to individuals with disabilities under Section 508 of the Rehabilitation Act, WCAG 2.1, or other applicable law. Exceptions apply for: (1) private non-commercial content not shared publicly, (2) AI-generated content used solely for internal business research not intended for public distribution, and (3) content using only minimal AI assistance that does not substantially alter meaning or authorship.

News media entities that publish generative AI content must (1) prominently display a label indicating the content is generative AI, (2) credit any source used to produce the content, and (3) include a disclaimer that the content may not accurately reflect the source material.

Sellers and sellers' agents who use generative AI or photo-editing software to add furniture or non-fixed items to dwelling photos must (1) disclose the use of AI or editing software in each advertisement, and (2) upon request, provide the original unaltered images to a buyer or buyer's agent.

News media entities that use generative AI content must (1) prominently display a label indicating the content is generative AI, (2) credit any source used to produce the content, and (3) include a disclaimer that the content may not accurately reflect the source material.

Covered providers must offer users the option to include a manifest (human-visible) disclosure in AI-generated content that identifies the content as generated by artificial intelligence, is clear, conspicuous, appropriate for the medium, understandable to a reasonable person, and permanent or extraordinarily difficult to remove.

Any person who distributes or publishes AI-generated content depicting a historically recognized human disaster and has actual knowledge the content is materially deceptive must include a disclosure label stating This (image, video, or audio) has been manipulated — printed in legible text for visual media and spoken at the beginning, end, and at two-minute intervals for audio content. Exemptions apply for satire/parody, bona fide news reporting with authenticity disclosures, broadcast stations with consistent disclaimer requirements, and initial platform dissemination of third-party content.

Publishers must conspicuously disclose on the cover of any book published in New York that was wholly or partially created through the use of generative artificial intelligence that the book was created with the use of generative artificial intelligence. This applies to all printed and digital books regardless of target age group or audience, including those consisting of text, pictures, audio, puzzles, games, or any combination thereof.

Any news media content published, broadcast, or otherwise disseminated or accessible in New York that was substantially composed, authored, or otherwise created using generative AI must carry a conspicuous disclosure stating the content was substantially created by generative AI. For written or visual content, the disclosure must be imprinted at the top of the page, webpage, image, graphic, or video. For audio content, the disclosure must be verbally stated at the onset. This disclosure requirement does not apply if the content is eligible for copyright registration.

Search engines must, when displaying information generated by artificial intelligence, inform the user that the information was AI-generated by providing two simultaneous disclosures: (1) a plain-language label in the same font size as the AI-generated information, placed directly above it, and (2) a watermark across the AI-generated information in the same font size and in clear, plain language.

Publishers of any book published in New York that was wholly or partially created through the use of generative artificial intelligence must conspicuously disclose on the cover of the book that it was created with the use of generative artificial intelligence. This applies to all printed and digital books regardless of target age group or audience, including books consisting of text, pictures, audio, puzzles, games, or any combination thereof.

Any party serving a paper or filing in a civil action that was drafted with the assistance of generative artificial intelligence must attach a separate affidavit disclosing such use and certifying that a human being has reviewed the source material and verified that the AI-generated content is accurate.

An appellant's brief must include, as a required component, a disclosure of any use of generative artificial intelligence in its drafting and a certification that the content was reviewed and verified by a human, where Rule 2107 applies.

Publishers of newspapers, magazines, or other publications printed or electronically published in New York must conspicuously imprint at the top of the page or webpage a disclosure that any article, periodical, photograph, video, or other visual image was composed through the use of artificial intelligence or other information communication technology, whenever the content was wholly or partially composed or authored using generative AI.

Any party serving a paper or file drafted with the assistance of generative artificial intelligence must attach a separate affidavit disclosing such use and certifying that a human reviewed the source material and verified the AI-generated content is accurate.

Appellants whose briefs were drafted with the assistance of generative artificial intelligence must include in the brief a disclosure of AI use and a certification that the content was reviewed and verified by a human.

Developers must program AI systems to produce a distinctive watermark on any AI-generated product that informs the user the content was generated using an artificial intelligence system.

No person, business, or organization may distribute for public or commercial use any AI-generated product that does not include a distinctive watermark indicating the content was generated using an artificial intelligence system. This obligation does not apply to AI-generated products created at a user's prompting solely for personal, noncommercial use.

Covered providers must offer each user the option to include a clear, conspicuous, medium-appropriate manifest disclosure identifying the content as AI-generated that is permanent or not reasonably removable to the extent technically feasible.

Digital platforms must attach a clear, conspicuous, medium-appropriate disclosure to any AI-simulation content identifying it as AI-generated and conveying (directly or via link) the simulation's name and version, the creation timestamp, and a unique identifier, consistent with industry standards and permanent or not reasonably removable to the extent technically feasible.

Any individual who creates or distributes an AI-generated image, text, simulation, or video must place a watermark on 30% of the content at minimum 50% opacity, containing the statement Artificial Intelligence Generated Material.

Any person or entity that uploads, creates, distributes, or originates AI-generated video, image, or photography on a public platform viewable in Rhode Island must include a visible marking disclosing that the content was generated by AI. Violations committed for profit or trade constitute deceptive trade practices subject to civil penalties under § 6-13.1-8.

Developers of AI systems that generate or manipulate synthetic digital content must ensure outputs are marked and detectable as synthetic before consumers first interact with or are exposed to them, in a consumer-detectable and accessibility-compliant manner, using technical solutions that are effective, interoperable, robust, and reliable consistent with recognized technical standards. Exemptions apply for text-only content, public-interest informational content, content unlikely to mislead a reasonable person, assistive editing, and compliance-detection tools.

Publishers of synthetic media must include a conspicuous disclosure stating that the content has been manipulated with AI or digital technology and is not an accurate representation of the depicted individual or event. For visual media, the disclosure must appear in text no smaller than the largest font in the media and, for video, for the full duration. For audio-only media, the disclosure must be clearly spoken at the beginning, end, and at intervals of no more than two minutes.

Federal officials must include a clear, conspicuous, plain-language disclaimer on any publicly released content created or manipulated using generative AI, stating that the content is AI-generated, explaining how it was generated or altered, and identifying the technology or method used. Exceptions apply for non-public communications, classified content (if an unclassified summary is retained), minor graphic adjustments, routine text drafts reviewed by agency staff before publication, and personal non-government social media posts.

Providers of generative AI systems that produce covered AI-generated content must label that content with a clear, conspicuous, embedded, and (where feasible) accessibility-conscious disclosure identifying it as AI-generated.

Any official police report or law-enforcement record created in whole or in part using generative AI must: (1) include a disclaimer that the report or record contains content generated by artificial intelligence; (2) where technically feasible, identify the specific content that was generated by AI; and (3) include a certification by the author that the author has read and reviewed the report or record for accuracy.

Developers must ensure that any generative AI system producing audio, image, or video content includes (1) a clear and conspicuous human-perceptible disclosure identifying the content as AI-generated and (2) embedded metadata identifying the content as AI-generated, the generating tool, and the creation date/time. Disclosures must be, to the extent technically feasible, permanent or not easily removable.

Developers must ensure that any generative AI system producing text content, including through an AI chatbot, includes a clear and conspicuous disclosure identifying the content as AI-generated that is, to the extent technically feasible, permanent or not easily removable.

Developers must ensure that generative AI systems producing audio, image, or video content include on each output (1) a clear and conspicuous human-perceptible label identifying the content as AI-generated and (2) embedded metadata identifying the content as AI-generated, the tool used, and the creation date and time. Disclosures must be permanent or unable to be easily removed.

Developers must ensure that generative AI systems producing text content — including through AI chatbots — include a clear and conspicuous disclosure identifying the content as AI-generated, which must be permanent or unable to be easily removed to the extent technically feasible.

Covered entities must ensure that any AI-generated media they create, distribute, publish, or transmit includes a clear and conspicuous human-perceptible disclosure — a watermark, text label, audio statement, or on-screen identifier — indicating the media was created or materially altered using AI. Visual media must include a watermark or text label; audio media must include a spoken or transcript statement at the beginning; video media must include an on-screen disclosure of at least 3 seconds at the start and a persistent watermark throughout playback.

GenAI hosting platforms must not knowingly make available a GenAI system that does not embed the latent provenance disclosures required by Section 22757.3.

Capture device manufacturers must, for any device first produced for sale in California on or after January 1, 2028, (1) embed latent provenance disclosures by default in captured content identifying the manufacturer name, device name and version, and creation or alteration timestamp, and (2) provide users with the option to include such disclosures. Compliance is required only to the extent technically feasible and compliant with widely adopted standards-body specifications.

Platforms that host GenAI systems for download (source code or model weights) must not knowingly make available any GenAI system that fails to include the latent provenance disclosures required by § 22757.3 (the existing covered-provider disclosure obligations). This effectively extends enforcement upstream: hosting platforms become gatekeepers that must verify their hosted GenAI systems embed proper provenance data before distribution. The 'knowingly' standard provides a scienter requirement — hosting platforms are not strictly liable for every non-compliant system but must not distribute them with actual knowledge of non-compliance.

Beginning January 1, 2028, manufacturers of cameras, phones, voice recorders, and similar capture devices must — for any device first produced for sale in California on or after that date — (1) give users the option to embed latent (machine-readable, not human-visible) provenance disclosures in captured content, conveying the manufacturer name, device name and version, and creation/alteration timestamp; and (2) enable this disclosure feature by default. Both obligations are subject to two safety valves: technical feasibility and compliance with widely adopted standards-body specifications. Assembly-only firms are excluded from the manufacturer definition. This is a notable extension of AI provenance law into hardware, aimed at creating an authenticity baseline for non-AI-generated content.

Covered providers must, to the extent commercially and technically reasonable, embed provenance data in AI-generated or materially altered audio, image, or video content enabling consumers to assess AI origin, and must use commercially reasonable tamper-resistant methods — including C2PA standards — to protect provenance data from removal or disassociation. Exemptions apply for B2B use, video games, upscaling-only systems, and trade secrets.

Capture device manufacturers must embed latent provenance disclosures in captured content identifying the manufacturer or providing digital signatures proving capture-device origin, plus creation/alteration timestamps, to the extent technically feasible and compliant with established standards-body specifications. Applies to devices produced for sale in Utah on or after January 1, 2028.

Covered providers must embed a latent, standards-compliant provenance disclosure in AI-generated or substantially AI-modified image, video, or audio content, conveying the creation/alteration timestamp and either the provider's name or digital signatures proving generative AI origin.

Developers of high-risk generative AI systems must ensure that synthetic content outputs are identifiable and detectable using industry-standard tools or developer-provided tools, with identification applied at the time of generation. For artistic, creative, satirical, or fictional works, identification must not hinder display or enjoyment. Text-only outputs, public-interest informational content, content unlikely to mislead a reasonable person, and assistive editing outputs are exempt.

Covered providers must, to the extent commercially and technically reasonable, embed provenance data in any video, image, or audio content (or any combination thereof) that is created or materially altered by the covered provider's generative AI system. The provenance data must allow a user to assess whether the content was created or materially altered by that system. Materially altered means a significant change that substantially alters the data; minor modifications such as brightness, contrast, color, sharpening, saturation, filters, resizing, scaling, cropping, format conversions, resampling, denoising, and background noise removal are excluded. Provenance data may not be required to include information relating to an identified or reasonably identifiable individual.

Covered providers must use commercially and technically reasonable methods to make provenance data difficult to remove or tamper with. Use of a commonly supported technical standard for watermarking or metadata, such as the Coalition for Content Provenance and Authenticity (C2PA) specification, is deemed compliant with this requirement.

Covered providers must embed a latent disclosure in all AI-generated image, video, or audio content created by their GenAI system that conveys (to the extent technically feasible) the provider's name, GenAI system name and version, creation or alteration timestamp, a unique identifier, and whether the content was generated or modified by AI. The disclosure must be permanent or extraordinarily difficult to remove, detectable by the provider's own tool, and compliant or interoperable with widely accepted industry standards.

Covered providers that license their GenAI system to a third party must contractually require the licensee to ensure the system includes provenance data meeting the latent disclosure criteria of § 22757.3(a), to the extent technically feasible.

Covered providers must revoke a third-party licensee's license within 48 hours of discovering that the licensee modified the licensed GenAI system so that it can no longer include the required latent disclosure in generated content.

Third-party licensees must cease using a licensed GenAI system after the covered provider revokes the license for failure to maintain required latent disclosures.

Synthetic content creations system providers must embed provenance data — conforming to C2PA Content Credentials or equivalent standards, cryptographically bound to the file — in all synthetic content their systems produce or modify. The provenance data must at minimum identify the content as synthetic, state that AI was used, name the provider, include a timestamp, identify AI-generated portions, and identify the generating device, system, or service.

State agencies must ensure, to the extent practicable, that all audio, images, and videos they publish or distribute electronically carry provenance data identifying — at minimum — the generating device/system/service, any AI-generated portions, whether AI was used, the AI provider name if applicable, and a timestamp.

Developers must include a clear and conspicuous disclosure identifying output from any generative AI system producing images, video, or audiovisual content as AI-generated content. The disclosure must include a human-perceptible label and embedded metadata identifying the content as AI-generated, the tool used, and the creation timestamp. The disclosure must, to the extent technically feasible, be permanent or unable to be easily removed. Visual-only content requires visual disclosure; audiovisual content requires both visual and audible disclosure. All disclosures must stand out by size, contrast, location, and duration; be unavoidable; use plain language; and not be contradicted by the content itself.

Generative AI tool providers must apply standards-compliant provenance data — either directly or via third-party technology — to all wholly-generated synthetic content their tool produces.

Generative AI tool providers must make available a provenance application tool that lets users apply provenance data to content significantly modified to include synthetic content.

Capture device manufacturers (cameras, phones with cameras, microphones, voice recorders) must include functionality allowing users to enable the inclusion of provenance data in their captured content.

Covered AI tool providers must embed a machine-readable provenance label in every image, video, or audio content instance created by their AI. The provenance label must: (1) be readable by the provider's provenance label reading tool; (2) be, to the extent technically feasible, permanent or extraordinarily difficult to remove; and (3) convey, to the extent technically feasible, the following system provenance data — either directly or via a link to a permanent website: (A) the name of the covered AI tool provider, (B) the name and version number of the AI that created or altered the content, (C) the time and date of the content's creation or alteration, and (D) a unique identifier of the content.

Capture device manufacturers must, for any capture device first produced for sale in Illinois on or after January 1, 2027, and to the extent technically feasible and compliant with widely adopted standards-body specifications: (1) provide users with the option to include a provenance label conveying system provenance data — manufacturer name, device name and version, and creation timestamp; (2) embed the provenance label in captured content by default; (3) clearly inform users of provenance label settings upon first use of a recording function; (4) provide a clear and accessible opt-out mechanism in device settings; and (5) ensure these provenance capabilities are available for the default capture application and for third-party applications that use the device's capture functionalities.

Covered AI tool providers that license their AI to third parties must require by contract that the licensee maintain the system's capability to include provenance labels. If the provider obtains actual knowledge that a third-party licensee has modified the AI to remove its provenance labeling capability, the provider must revoke the licensee's license within 96 hours.

A third-party licensee whose license to use AI has been revoked under this Act must immediately cease using the AI.

Operators of websites or applications that make AI source code or model weights available for download must not knowingly make available AI that does not include provenance labels in generated content as required by Section 10(f).

Generative AI systems used within Massachusetts must include on or within AI-generated content (1) a clear and conspicuous human-perceptible notice identifying the content as AI-generated, permanent or difficult to remove to the extent technically feasible, and (2) embedded metadata identifying the content as AI-generated, the identity of the generating system, and the date and time of creation.

Any person within Massachusetts who uses a generative AI system to create or repurpose AI-generated content must not remove any AI-disclosure notice or metadata required by Section 2.

GenAI Providers must embed standards-compliant Provenance Data in all wholly AI-generated Synthetic Content produced by their GenAI Tools, either directly or through third-party technology.

GenAI Providers must make available a Provenance Application Tool that enables users to apply standards-compliant Provenance Data to content that has been modified to include Synthetic Content.

Capture Device manufacturers must (1) include in the device's default capture app the ability for users to enable inclusion of Provenance Data in captured content and (2) ensure secure hardware-based provenance capture is available to third-party applications.

Persons responsible for AI-generated political communications must embed in the communication's metadata the disclosure statement, the identity of the generative AI tool used, and the date and time the communication was created.

Persons responsible for AI-generated political communications must ensure that the disclosure statement — including the metadata disclosure — is, to the extent technically feasible, permanent or unable to be easily removed by subsequent users.

Covered providers must embed a latent (machine-readable) disclosure by default in all synthetic content that conveys the provider name, system name and version, creation/alteration timestamp, and a unique content identifier; is detectable by a provenance detection tool; complies with provenance standards accepted by an established standards-setting body; and is permanent or extraordinarily difficult to remove.

Covered providers must contractually require third-party licensees of their generative AI system to continue complying with the disclosure requirements of Section 3, and must revoke the license within 96 hours of discovering material noncompliance.

Capture device manufacturers must, for devices first produced for sale in the state on or after January 1, 2028, embed a latent provenance disclosure by default identifying the manufacturer, device model and version, and date and time of capture, and must provide users the option to opt out. Compliance must be aligned with established standards-body standards to the extent technically feasible.

Campaigns for the offices of Governor, Lieutenant Governor, Attorney General, or Comptroller must, beginning with the 2030 election cycle, embed C2PA-standard provenance data in all political communications that include images or videos, disclosing: the type of device or system used to generate the content, any synthetic content portions, whether AI was used to create or edit the content, the name of the generative AI provider (if any), and the timestamp of provenance data application.

Covered providers must embed a latent (machine-readable) provenance disclosure in AI-generated image, video, and audio content conveying the provider name, system name and version, creation timestamp, and a unique identifier; the signal must be detectable by the provider's detection tool, consistent with widely accepted industry standards, and permanent or difficult to remove.

Covered providers must contractually require licensees of their generative AI system to preserve the latent-disclosure capability, revoke a license within four calendar days of learning a licensee disabled it, and licensees must stop using the system after revocation.

GenAI system hosting platforms must not knowingly make available for download any generative AI system that fails to place the manifest and latent disclosures required by Section 4.

Digital platforms must attach a clear, conspicuous, medium-appropriate disclosure to any AI-simulation content identifying it as AI-generated and conveying (directly or via link) the simulation's name and version, the creation timestamp, and a unique identifier, consistent with industry standards and permanent or not reasonably removable to the extent technically feasible.

Covered AI tool providers must embed a permanent or extraordinarily-difficult-to-remove, machine-readable provenance label in every AI-generated image, video, and audio instance, conveying the provider name, system name and version, creation timestamp, and the type and degree of AI augmentation.

Capture-device manufacturers must, for devices produced for sale in Pennsylvania, embed provenance labels by default (conveying manufacturer name, device name and version, and creation timestamp), offer a user opt-out, inform users of provenance settings on first use, and display a clear indicator when a label is applied.

Covered AI tool providers licensing their generative system to a third party must contractually require the licensee to maintain the system's provenance-labeling capability.

Covered AI tool providers must, within 96 hours of obtaining actual knowledge that a licensee has modified the system to remove its provenance-label capability, revoke that licensee's license; the licensee may not use the system after revocation.

Developers of AI systems that generate or manipulate synthetic digital content must ensure outputs are marked and detectable as synthetic before consumers first interact with or are exposed to them, in a consumer-detectable and accessibility-compliant manner, using technical solutions that are effective, interoperable, robust, and reliable consistent with recognized technical standards. Exemptions apply for text-only content, public-interest informational content, content unlikely to mislead a reasonable person, assistive editing, and compliance-detection tools.

Providers of generative AI applications must ensure that audio or visual content created or substantially modified by their application incorporates a machine-readable disclosure acknowledging the generative AI origin of such content, and must establish and implement reasonable measures to prevent the disclosure from being removed or tampered with. Compliance with FTC-approved self-regulatory guidelines satisfies this requirement.

Providers of generative AI applications must make available to users the ability to incorporate, within the metadata of AI-generated or AI-modified content, tamper-evident provenance information including the application name, AI model name and version, creation/modification date and time, and the portion of content created or modified.

Persons who commercially offer tools primarily used to create synthetic content or synthetically-modified content must (1) provide users the ability to include content provenance information indicating AI origin, and (2) when a user opts to include such information, implement reasonable security measures to ensure it is machine-readable and not easily removed, altered, or separated from the content.

Persons who commercially offer tools primarily used to create or substantially modify covered content (copyrighted works) must (1) provide users the ability to include content provenance information for any covered content created or modified by the tool, and (2) when a user opts to include such information, implement reasonable security measures to ensure it is machine-readable and not easily removed, altered, or separated from the content.

Providers must bind or embed a machine-readable provenance disclosure in covered AI-generated content identifying the content, the system and version used, the creation or modification timestamp, and conforming to interoperable standards set by the FTC and Working Group, without including the user's personally identifiable information.

Developers must embed standards-compliant provenance data in all synthetic digital content wholly generated by their generative AI system or service, either directly or through third-party technologies.

Developers must ensure that any generative AI system producing audio, image, or video content includes (1) a clear and conspicuous human-perceptible disclosure identifying the content as AI-generated and (2) embedded metadata identifying the content as AI-generated, the generating tool, and the creation date/time. Disclosures must be, to the extent technically feasible, permanent or not easily removable.

Developers must ensure that generative AI systems producing audio, image, or video content include on each output (1) a clear and conspicuous human-perceptible label identifying the content as AI-generated and (2) embedded metadata identifying the content as AI-generated, the tool used, and the creation date and time. Disclosures must be permanent or unable to be easily removed.

Developers must embed standards-compliant provenance data — identifying whether content is synthetic and which developer's system generated it — into all synthetic digital content wholly generated by their generative AI system or service, either directly or through third-party technologies.

Capture device manufacturers must include a provenance application tool that is made known and available to the user, enabled by default, and usable to apply provenance data to digital content captured with the device.

Developers of high-risk generative AI systems that generate or substantially modify synthetic content must ensure that outputs are (1) identifiable and detectable by consumers using industry-standard tools or tools provided by the developer, (2) compliant with applicable accessibility requirements to the extent reasonably feasible, and (3) identified at the time the output is generated. For audio, image, or video content that is part of an evidently artistic, creative, satirical, fictional, or analogous work, identification must be applied in a manner that does not hinder display or enjoyment. Exemptions apply to text-only content published in the public interest or unlikely to mislead a reasonable person, and to outputs of systems performing assistive standard editing, not substantially altering input data, or used for authorized crime detection/prosecution.

Developers of high-risk generative AI systems that generate or substantially modify synthetic content must ensure that outputs are: (1) identifiable and detectable using industry-standard tools or tools provided by the developer; (2) compliant with applicable accessibility requirements, to the extent reasonably feasible; and (3) identified at the time of generation. For audio, image, or video content that is part of an evidently artistic, creative, satirical, fictional, or analogous work, identification must not hinder display or enjoyment. The identification requirement does not apply to: (i) synthetic content consisting exclusively of text, content published to inform the public on a matter of public interest, or content unlikely to mislead a reasonable person; or (ii) outputs of a system performing an assistive editing function, not substantially altering input data, or used for law enforcement purposes.

Capture device manufacturers must, for any device first produced for sale in California on or after January 1, 2028, (1) embed latent provenance disclosures by default in captured content identifying the manufacturer name, device name and version, and creation or alteration timestamp, and (2) provide users with the option to include such disclosures. Compliance is required only to the extent technically feasible and compliant with widely adopted standards-body specifications.

Platforms that host GenAI systems for download (source code or model weights) must not knowingly make available any GenAI system that fails to include the latent provenance disclosures required by § 22757.3 (the existing covered-provider disclosure obligations). This effectively extends enforcement upstream: hosting platforms become gatekeepers that must verify their hosted GenAI systems embed proper provenance data before distribution. The 'knowingly' standard provides a scienter requirement — hosting platforms are not strictly liable for every non-compliant system but must not distribute them with actual knowledge of non-compliance.

Beginning January 1, 2028, manufacturers of cameras, phones, voice recorders, and similar capture devices must — for any device first produced for sale in California on or after that date — (1) give users the option to embed latent (machine-readable, not human-visible) provenance disclosures in captured content, conveying the manufacturer name, device name and version, and creation/alteration timestamp; and (2) enable this disclosure feature by default. Both obligations are subject to two safety valves: technical feasibility and compliance with widely adopted standards-body specifications. Assembly-only firms are excluded from the manufacturer definition. This is a notable extension of AI provenance law into hardware, aimed at creating an authenticity baseline for non-AI-generated content.

Covered providers must, to the extent commercially and technically reasonable, embed provenance data in AI-generated or materially altered audio, image, or video content enabling consumers to assess AI origin, and must use commercially reasonable tamper-resistant methods — including C2PA standards — to protect provenance data from removal or disassociation. Exemptions apply for B2B use, video games, upscaling-only systems, and trade secrets.

Capture device manufacturers must embed latent provenance disclosures in captured content identifying the manufacturer or providing digital signatures proving capture-device origin, plus creation/alteration timestamps, to the extent technically feasible and compliant with established standards-body specifications. Applies to devices produced for sale in Utah on or after January 1, 2028.

Covered providers must embed a latent, standards-compliant provenance disclosure in AI-generated or substantially AI-modified image, video, or audio content, conveying the creation/alteration timestamp and either the provider's name or digital signatures proving generative AI origin.

Covered providers must embed a latent disclosure in all AI-generated image, video, or audio content created by their GenAI system that conveys (to the extent technically feasible) the provider's name, GenAI system name and version, creation or alteration timestamp, a unique identifier, and whether the content was generated or modified by AI. The disclosure must be permanent or extraordinarily difficult to remove, detectable by the provider's own tool, and compliant or interoperable with widely accepted industry standards.

GenAI Providers must embed standards-compliant Provenance Data in all wholly AI-generated Synthetic Content produced by their GenAI Tools, either directly or through third-party technology.

Covered providers must embed a latent (machine-readable) disclosure by default in all synthetic content that conveys the provider name, system name and version, creation/alteration timestamp, and a unique content identifier; is detectable by a provenance detection tool; complies with provenance standards accepted by an established standards-setting body; and is permanent or extraordinarily difficult to remove.

Capture device manufacturers must, for devices first produced for sale in the state on or after January 1, 2028, embed a latent provenance disclosure by default identifying the manufacturer, device model and version, and date and time of capture, and must provide users the option to opt out. Compliance must be aligned with established standards-body standards to the extent technically feasible.

Covered providers must embed a latent (machine-readable) provenance disclosure in AI-generated image, video, and audio content conveying the provider name, system name and version, creation timestamp, and a unique identifier; the signal must be detectable by the provider's detection tool, consistent with widely accepted industry standards, and permanent or difficult to remove.

Digital platforms must attach a clear, conspicuous, medium-appropriate disclosure to any AI-simulation content identifying it as AI-generated and conveying (directly or via link) the simulation's name and version, the creation timestamp, and a unique identifier, consistent with industry standards and permanent or not reasonably removable to the extent technically feasible.

Developers of AI systems that generate or manipulate synthetic digital content must ensure outputs are marked and detectable as synthetic before consumers first interact with or are exposed to them, in a consumer-detectable and accessibility-compliant manner, using technical solutions that are effective, interoperable, robust, and reliable consistent with recognized technical standards. Exemptions apply for text-only content, public-interest informational content, content unlikely to mislead a reasonable person, assistive editing, and compliance-detection tools.

Providers of generative AI applications must collaborate with providers of covered online platforms to assist those platforms in identifying and accessing the embedded provenance disclosures.

NIST (Under Secretary) must establish a public-private partnership to develop voluntary, consensus-based standards for content provenance information, watermarking, and synthetic content detection across images, audio, video, text, and multimodal content, in consultation with the Register of Copyrights and the USPTO Director.

Providers must bind or embed a machine-readable provenance disclosure in covered AI-generated content identifying the content, the system and version used, the creation or modification timestamp, and conforming to interoperable standards set by the FTC and Working Group, without including the user's personally identifiable information.

Developers must embed standards-compliant provenance data in all synthetic digital content wholly generated by their generative AI system or service, either directly or through third-party technologies.

Developers must embed standards-compliant provenance data — identifying whether content is synthetic and which developer's system generated it — into all synthetic digital content wholly generated by their generative AI system or service, either directly or through third-party technologies.

Large online platforms must detect whether any standards-compliant provenance data is embedded into or attached to content distributed on the platform.

Large online platforms (social media, file-sharing, mass messaging platforms, and stand-alone search engines with 2M+ unique monthly users) must scan content distributed on their platform to detect any provenance data that conforms to widely adopted standards-body specifications. Where system provenance data is found indicating AI generation, substantial AI alteration, or capture-device origin, the platform must provide a user interface that clearly and conspicuously discloses: whether provenance data exists, the name of the GenAI system or capture device that created or altered the content, and whether digital signatures are available. This is a detect-and-display obligation — the platform need only surface provenance signals that are already embedded in content using recognized standards.

Large online platforms must (1) detect whether compliant system provenance data is embedded in or attached to content distributed on the platform, (2) provide a user interface disclosing the availability of system provenance data, and (3) allow users to inspect all available compliant system provenance data through the platform interface, download, or a link.

Large online platforms must, to the extent technically feasible: (1) detect whether any provenance label compliant with widely adopted standards-body specifications is embedded in or attached to content distributed on the platform; (2) provide a clear and conspicuous mechanism disclosing to users that provenance data is available for detected content; and (3) allow users to inspect all available system provenance data in an easily accessible manner — either directly through the platform's user interface or by enabling the user to download the content with its attached system provenance data.

Large Online Platforms must (1) retain any available Provenance Data in content provided to or posted on the platform and (2) make available to content consumers either the Provenance Data itself, a conspicuous indicator that Provenance Data is available, or both.

Large online platforms must (1) detect system provenance data in disseminated content, (2) append the platform name and posting date to system provenance data, (3) provide a user interface disclosing the available system provenance data, and (4) allow users to inspect, download, or otherwise access system provenance data.

Large online platforms must detect standards-compliant provenance data in distributed content and provide a user interface that conspicuously discloses whether provenance is available, the name of the generating AI system, and whether digital signatures exist, and that lets users inspect or download the provenance data.

Large online platforms must detect machine-readable provenance labels in content they distribute and provide a conspicuous user-facing mechanism that discloses whether a provenance label is available and lets users inspect it.

Covered platforms must implement and maintain commercially reasonable policies and technical measures designed to detect synthetic media that materially depicts real individuals, statements, or events.

Covered online platforms must (1) conspicuously identify displayed content carrying a machine-readable AI disclosure, (2) not tamper with or strip such disclosures, including on transfer, (3) offer users a provenance-sharing option, (4) make a good-faith effort to combat the liar's dividend using FTC-recommended strategies, and (5) keep the information accessible where feasible.

Controllers of online services, products, or features must (1) retain any available provenance data in digital content provided to their platform and (2) make available to users either the provenance data, a conspicuous indicator of whether provenance data is available, or both.

Platforms hosting or distributing AI-generated media must (1) maintain functionality to preserve disclosures and watermarks, (2) prohibit removal or obfuscation of disclosures without explicit lawful justification, and (3) implement automated detection systems, when feasible, to identify unlabeled AI-generated content and prompt disclosure.

Large online platforms must provide a user interface that clearly and conspicuously discloses when content carries system provenance data indicating AI generation or device capture, including whether provenance data is available, the name of the generating GenAI system or capture device, and whether digital signatures are available. Platforms must also allow users to inspect all available standards-compliant system provenance data through the platform UI, downloadable content, or a link to a third-party display.

Large online platforms (social media, file-sharing, mass messaging platforms, and stand-alone search engines with 2M+ unique monthly users) must scan content distributed on their platform to detect any provenance data that conforms to widely adopted standards-body specifications. Where system provenance data is found indicating AI generation, substantial AI alteration, or capture-device origin, the platform must provide a user interface that clearly and conspicuously discloses: whether provenance data exists, the name of the GenAI system or capture device that created or altered the content, and whether digital signatures are available. This is a detect-and-display obligation — the platform need only surface provenance signals that are already embedded in content using recognized standards.

Beyond detecting and disclosing provenance data, large online platforms must allow users to inspect the full system provenance data in an easily accessible way. The platform may satisfy this through any of three methods: (1) inline display through its own UI, (2) enabling the user to download the content with provenance data attached, or (3) providing a link to the provenance data on a website or app (the platform's own or a third party's). Platforms have flexibility in which method to use but must offer at least one.

Large online platforms must (1) detect whether compliant system provenance data is embedded in or attached to content distributed on the platform, (2) provide a user interface disclosing the availability of system provenance data, and (3) allow users to inspect all available compliant system provenance data through the platform interface, download, or a link.

Large online platforms (≥1,000,000 monthly users) must (1) retain any provenance data in content posted on or provided to the platform, and (2) make either the provenance data or a conspicuous indicator that provenance data is available accessible to Illinois consumers.

Large online platforms must, to the extent technically feasible: (1) detect whether any provenance label compliant with widely adopted standards-body specifications is embedded in or attached to content distributed on the platform; (2) provide a clear and conspicuous mechanism disclosing to users that provenance data is available for detected content; and (3) allow users to inspect all available system provenance data in an easily accessible manner — either directly through the platform's user interface or by enabling the user to download the content with its attached system provenance data.

Large Online Platforms must (1) retain any available Provenance Data in content provided to or posted on the platform and (2) make available to content consumers either the Provenance Data itself, a conspicuous indicator that Provenance Data is available, or both.

Large online platforms must (1) detect system provenance data in disseminated content, (2) append the platform name and posting date to system provenance data, (3) provide a user interface disclosing the available system provenance data, and (4) allow users to inspect, download, or otherwise access system provenance data.

Online platform operators must provide or allow users to use labeling tools sufficient to label materially deceptive media as required by § 1711, and must not lock or otherwise impede the use of those labeling tools.

Large online platforms must detect standards-compliant provenance data in distributed content and provide a user interface that conspicuously discloses whether provenance is available, the name of the generating AI system, and whether digital signatures exist, and that lets users inspect or download the provenance data.

Large online platforms must detect machine-readable provenance labels in content they distribute and provide a conspicuous user-facing mechanism that discloses whether a provenance label is available and lets users inspect it.

Covered platforms must clearly and conspicuously disclose that content has been generated or materially altered using artificial intelligence upon actual knowledge or reasonable detection of synthetic media.

Providers of covered online platforms must clearly and conspicuously disclose to users, for any audio or visual content that incorporates an AI-origin provenance disclosure, the information included in that disclosure.

Covered online platforms must (1) conspicuously identify displayed content carrying a machine-readable AI disclosure, (2) not tamper with or strip such disclosures, including on transfer, (3) offer users a provenance-sharing option, (4) make a good-faith effort to combat the liar's dividend using FTC-recommended strategies, and (5) keep the information accessible where feasible.

Controllers of online services, products, or features must (1) retain any available provenance data in digital content provided to their platform and (2) make available to users either the provenance data, a conspicuous indicator of whether provenance data is available, or both.

Controllers of online services must (1) retain any available provenance data in digital content provided to the service and not strip it, and (2) make available to users either the provenance data, a conspicuous indicator of whether provenance data is available, or both.

Large online platforms must not knowingly strip standards-compliant system provenance data or digital signatures from content uploaded or distributed on the platform, to the extent technically feasible.

Large online platforms must not knowingly strip standards-compliant system provenance data or digital signatures from content that is uploaded to or distributed on the platform, to the extent this is technically feasible. This is a preservation duty — the platform need not add provenance data, but it must not remove what is already there. The 'knowingly' and 'technically feasible' qualifiers provide a safe harbor for incidental or unavoidable data loss during normal processing, but deliberate removal of recognized provenance signals is prohibited.

Large online platforms must not, to the extent technically feasible, knowingly strip standards-compliant system provenance data or digital signatures from content uploaded or distributed on the platform.

An advertising medium must not transmit, distribute, display, air, or otherwise make available an advertisement containing a synthetic performer after both of the following conditions are met: (1) a court of competent jurisdiction has issued an order finding that the advertisement violates this chapter or enjoining the creator from publishing the advertisement, and (2) the advertising medium has been served with the order and information reasonably sufficient to identify the advertisement. Upon receipt of such an order, the advertising medium must, as soon as commercially reasonable and technically feasible, remove or disable access to the advertisement in California and cease accepting payment for its further dissemination.

Synthetic content creations system hosting platforms must not make available any system whose provider does not apply compliant provenance data, and must not deliberately prevent a provider from applying provenance data to content created or substantially modified by a synthetic content creations system.

Social media platforms must not delete, disassociate, or degrade provenance data from or associated with content uploaded by covered users, unless the content is permanently removed from the platform or removal of the provenance data is required by law.

Large online platforms (≥1,000,000 monthly users) must (1) retain any provenance data in content posted on or provided to the platform, and (2) make either the provenance data or a conspicuous indicator that provenance data is available accessible to Illinois consumers.

Large online platforms must not, to the extent technically feasible, knowingly strip any provenance label or system provenance data compliant with widely adopted standards-body specifications from content uploaded to or distributed on the platform.

Large Online Platforms must (1) retain any available Provenance Data in content provided to or posted on the platform and (2) make available to content consumers either the Provenance Data itself, a conspicuous indicator that Provenance Data is available, or both.

Large online platforms must not strip system provenance data or digital signatures that comply with standards set by an established standards-setting body.

Online platform operators must not strip AI provenance metadata from any content posted to their platform.

Large online platforms must not knowingly strip standards-compliant system provenance data or digital signatures from content uploaded or distributed on the platform.

Large online platforms must not strip provenance labels from content uploaded or distributed on the platform and must not retain personal provenance data from shared content.

Providers of covered online platforms must not remove any embedded provenance disclosure or incorporated metadata from audio or visual content distributed on the platform.

No person may knowingly remove, alter, tamper with, or disable content provenance information in furtherance of an unfair or deceptive act or practice in or affecting commerce.

Covered platforms must not remove, alter, tamper with, or disable content provenance information, or separate it from content so that users cannot access it. A narrow exception applies for security research that is necessary, proportionate, and limited to enhancing the platform's security.

Covered online platforms must (1) conspicuously identify displayed content carrying a machine-readable AI disclosure, (2) not tamper with or strip such disclosures, including on transfer, (3) offer users a provenance-sharing option, (4) make a good-faith effort to combat the liar's dividend using FTC-recommended strategies, and (5) keep the information accessible where feasible.

No person may knowingly falsify AI provenance disclosures, knowingly distribute for financial benefit AI content stripped of required disclosures or non-AI content bearing false disclosures, or manufacture, import, or offer tools primarily designed or marketed to circumvent, remove, tamper with, or falsely add such disclosures.

Controllers of online services, products, or features must (1) retain any available provenance data in digital content provided to their platform and (2) make available to users either the provenance data, a conspicuous indicator of whether provenance data is available, or both.

Controllers of online services must (1) retain any available provenance data in digital content provided to the service and not strip it, and (2) make available to users either the provenance data, a conspicuous indicator of whether provenance data is available, or both.

Platforms hosting or distributing AI-generated media must (1) maintain functionality to preserve disclosures and watermarks, (2) prohibit removal or obfuscation of disclosures without explicit lawful justification, and (3) implement automated detection systems, when feasible, to identify unlabeled AI-generated content and prompt disclosure.

Covered providers must make available, at no cost, a publicly accessible disclosure verification tool that allows users to assess whether image, video, or audio content was created or altered by the provider's GenAI system, outputs detected system provenance data, suppresses personal information absent express user consent, supports content upload and URL submission, and offers an API for programmatic access.

Covered providers must collect user feedback on the efficacy of their disclosure verification tool and incorporate relevant feedback into efforts to improve the tool.

Generative AI tool providers must make a provenance reader publicly available that allows any user to view provenance data and verify its integrity.

Covered AI tool providers must offer a free, publicly accessible provenance label reading tool that allows any person to upload content or submit a URL and determine whether the content was created or altered by the provider's AI. The tool must be accessible via a conspicuous link on the provider's website and mobile application, and must also support programmatic access through an API. The tool must include a feedback mechanism for users to submit efficacy feedback, and the provider must consider and use that feedback to improve the tool.

GenAI Providers must make a Provenance Reader publicly available that allows any person to view Provenance Data and verify its integrity.

Covered providers must make available at no cost a publicly accessible provenance detection tool that allows users to read embedded provenance data, reveals only system provenance data (not personal provenance data), supports uploads, URLs, and API integration, and complies with established standards-body standards.

Covered providers must not retain content or personal provenance data submitted to the provenance detection tool longer than reasonably necessary to perform the detection function.

Covered providers must offer, free of charge, a publicly accessible AI-detection tool (with an API) that accepts uploaded content or a URL, returns whether the content was created or altered by the provider's generative AI system, outputs detected system provenance data, and must collect and incorporate user feedback to improve the tool.

Covered AI tool providers must offer a free, publicly accessible provenance-label reading tool — linked conspicuously on their website and mobile app, accepting content by upload or URL, and supporting programmatic API submission — that lets anyone determine whether content was created or altered by the provider's generative AI system.

Providers must ensure users and covered platforms can detect that content is AI-generated and view the provenance data without undue financial burden, either by making it detectable by widely available tools or, if none exist, by providing a detection tool and clear instructions.

Developers must make publicly available (1) a provenance application tool enabling users to apply provenance data to content modified to include synthetic digital content, and (2) a provenance reader enabling users to view provenance data for synthetic digital content.

Developers must make publicly available (1) a provenance application tool enabling users to apply provenance data to synthetic digital content, and (2) a provenance reader enabling users to view provenance data of synthetic digital content.

Developers must implement reasonable downstream protection procedures to prevent use of the generative AI system without the required disclosures, including: (1) contractual provisions requiring end users and third-party licensees to refrain from removing disclosures; (2) certifications from end users and third-party licensees that they will not remove disclosures; and (3) termination of access when the developer has reason to believe a disclosure has been removed.

Third-party licensees must implement reasonable downstream protection procedures to prevent use of the generative AI system without the required disclosures, including: (1) contractual provisions requiring end users to refrain from removing disclosures; (2) certifications from end users that they will not remove disclosures; and (3) termination of access when there is reason to believe an end user has removed a required disclosure.

Providers must collaborate with covered online platforms to help them meet their disclosure-preservation obligations for content created or substantially modified by the provider's system.

Developers must implement reasonable procedures to prevent downstream use of AI-generated content without required disclosures, including (1) contractually requiring end users and third-party licensees to refrain from removing disclosures, (2) requiring certification of non-removal, and (3) terminating access when the developer has reason to believe a disclosure has been removed.

Third-party licensees must implement reasonable procedures to prevent downstream end-user use of the generative AI system without required disclosures, including (1) contractually requiring end users to refrain from removing disclosures, (2) requiring certification of non-removal, and (3) terminating access when there is reason to believe a disclosure has been removed.

Developers must implement reasonable procedures to prevent downstream use of the generative AI system without required disclosures, including (1) contractually requiring end users and third-party licensees to refrain from removing disclosures, (2) requiring certification of non-removal, and (3) terminating access when the developer has reason to believe a disclosure has been removed.

Third-party licensees must implement reasonable procedures to prevent downstream use of the generative AI system without required disclosures, including (1) contractually requiring end users to refrain from removing disclosures, (2) requiring certification of non-removal, and (3) terminating access when the developer has reason to believe a disclosure has been removed.