Washington AI Content Transparency Act

Overview

Washington House Bill 1170 (HB 1170), officially titled "AN ACT Relating to informing users when content is developed or modified by artificial intelligence; and adding a new chapter to Title 19 RCW," represents a significant legislative effort by the State of Washington to address the growing concerns surrounding artificial intelligence (AI) generated content. The primary objective of this legislation is to enhance transparency and provide consumers with the necessary tools and information to distinguish between authentic and AI-created or materially altered digital content. It specifically targets generative AI systems that are publicly accessible within Washington State and have a substantial user base, defined as over one million monthly visitors or users. The bill mandates a multi-faceted approach to disclosure, requiring covered providers to embed provenance data, offer AI detection tools, and provide options for manifest disclosures, while also stipulating latent disclosures within the content itself. This comprehensive framework aims to empower users and mitigate potential harms associated with deceptive or unidentifiable AI-generated media, such as deepfakes and misinformation.

The scope of HB 1170 extends to various forms of digital content, including video, image, and audio, and combinations thereof, that are either created or materially altered by generative AI systems. A key component of the bill is the requirement for "covered providers" to implement technical safeguards that make provenance data difficult to remove or tamper with, aligning with common industry standards like watermarking or metadata. Furthermore, the act includes provisions for government agencies, mandating that they clearly and conspicuously disclose to consumers when they are interacting with an AI system, regardless of whether the AI interaction would be obvious. While establishing robust transparency requirements, the bill also outlines specific exemptions, such as business-to-business uses, sales, licensing, or distribution of AI systems, content exclusively for video games or interactive experiences, and systems used solely for upscaling, noise reduction, or compression. This nuanced approach seeks to balance consumer protection with fostering innovation in the AI sector.

Definitions

House Bill 1170 establishes a set of key definitions crucial for its interpretation and application. "Artificial intelligence" or "AI" is defined as an engineered or machine-based system that exhibits varying levels of autonomy and can infer from input to generate outputs that influence physical or virtual environments, for explicit or implicit objectives. This broad definition encompasses the range of AI technologies the bill intends to regulate. A "covered provider" is specified as a person or entity that creates, codes, or otherwise produces a generative artificial intelligence system with over 1,000,000 monthly visitors or users, and which is publicly accessible within the geographic boundaries of Washington State. This threshold is designed to focus regulatory efforts on widely used AI systems.

The legislation further defines a "generative artificial intelligence system" as an AI capable of generating derived synthetic content, including text, images, video, and audio, that emulates the structure and characteristics of its training data. This definition directly targets the type of AI that can produce realistic and potentially deceptive content. Other important terms include "latent," meaning present but not manifest, and "manifest," referring to something easily perceived, understood, or recognized by a natural person. "Metadata" is defined as structural or descriptive information about data. The bill also distinguishes between "personal information," as defined in existing state law (RCW 19.255.005), and "personal provenance data," which is provenance data containing either personal information or unique device, system, or service information reasonably capable of being associated with a particular user. Crucially, "provenance data" itself is defined as data embedded into digital content or an AI system that verifies its authenticity or modification history. Finally, "materially altered" refers to significant changes that substantially modify the content's data, excluding minor edits like brightness adjustments or cropping, ensuring that only significant AI modifications fall under the bill's purview.

Governance and Institutional Framework

The governance and institutional framework for Washington HB 1170 primarily vests enforcement authority in the Washington State Attorney General. This legislative choice integrates the oversight of AI content transparency into existing consumer protection mechanisms, leveraging the Attorney General's established powers to address deceptive practices and ensure marketplace fairness. The bill explicitly states that the Attorney General can bring actions to enforce its provisions under the Consumer Protection Act (Chapter 19.86 RCW). This means that violations of HB 1170's requirements regarding AI-generated content disclosure and provenance data can be treated as unfair or deceptive acts or practices in commerce, subject to the remedies and penalties available under the Consumer Protection Act. This framework provides a clear avenue for accountability and redress for consumers impacted by non-compliant AI systems.

By placing enforcement under the Consumer Protection Act, the bill streamlines the regulatory process and avoids the creation of entirely new oversight bodies, which can often be resource-intensive and slow to establish. This approach also ensures that the enforcement efforts benefit from the Attorney General's expertise in investigating and litigating consumer fraud and deceptive advertising cases. The legislation's emphasis on transparency and non-deceptive practices aligns well with the core mission of consumer protection. Furthermore, the bill specifies that any waivers of its provisions are void, reinforcing the mandatory nature of its requirements and preventing covered providers from contractually circumventing their obligations. This institutional setup aims to provide a robust and efficient mechanism for upholding the integrity of digital content in an era increasingly influenced by artificial intelligence.

Key Focus Areas

Washington HB 1170 centers on several key focus areas to achieve its goal of AI training data transparency. A primary focus is the mandatory embedding of "provenance data" into digital content created or materially altered by generative AI systems. This data serves as an immutable record of the content's origin or modification history, designed to be difficult to remove or tamper with. The bill suggests that compliance with this requirement can be met through common standards such as watermarking or metadata, providing flexibility while ensuring technical robustness. This ensures that the digital footprint of AI involvement is preserved, allowing for later verification and identification. The objective is to provide an inherent mechanism for tracing the AI's role in content generation, thereby combating the spread of undetectable synthetic media.

Another critical area is the requirement for covered providers to offer free AI detection tools to users. These tools must enable users to assess whether content was created or altered by AI, provide system provenance data, and be accessible without requiring a visit to the provider's website. This empowers individual users to verify content authenticity. Complementing this, the bill mandates both "manifest disclosures" and "latent disclosures." Manifest disclosures are clear and conspicuous labels that identify content as AI-generated, designed to be perceptible to a normal consumer and extraordinarily difficult to remove. Latent disclosures, on the other hand, are embedded information (e.g., metadata) that is not perceptible to a normal consumer but is detectable by the provider's own AI detection tool. Furthermore, the legislation imposes specific disclosure requirements on government agencies, obliging them to clearly and conspicuously inform consumers when they are interacting with an AI system, using plain language and avoiding deceptive practices, irrespective of whether the AI interaction is obvious. These combined measures aim to create a comprehensive ecosystem of transparency for AI-generated content.

Implementation Framework

The implementation framework for Washington HB 1170 places significant responsibilities on "covered providers" of generative AI systems. These providers, defined by their substantial user base (over one million monthly visitors) and public accessibility within Washington State, are mandated to embed provenance data into any video, image, audio, or combination thereof that their AI systems create or materially alter. This embedding must utilize reasonable measures to protect the provenance data from removal or tampering, with common standards like watermarking or metadata being acceptable methods. This technical requirement forms the bedrock of the bill's transparency objectives, ensuring that the origins of AI-generated content are traceable and verifiable. The bill emphasizes that these technical safeguards should be robust enough to maintain the integrity of the provenance information.

Beyond embedding provenance data, the implementation framework requires covered providers to offer a free, publicly accessible AI detection tool. This tool must allow users to determine if content was generated or modified by AI and provide associated system provenance data, all without requiring a visit to the provider's website. Providers are also required to collect user feedback on the efficacy of these detection tools and incorporate relevant feedback into efforts to improve them, fostering a continuous improvement cycle. Furthermore, the bill stipulates that covered providers must offer users the option to include a "manifest disclosure"—a clear, conspicuous, and difficult-to-remove label identifying content as AI-generated—and must ensure that "latent disclosures" (imperceptible but detectable metadata) are present. For third-party licensees of generative AI systems, the bill mandates that covered providers require, by contract, that the licensee maintains the system's ability to include the necessary disclosures. If a covered provider discovers a licensee has modified a system to prevent such disclosures, the license must be revoked within 96 hours, and the licensee must cease using the system. These provisions collectively establish a detailed operational framework for compliance.

Monitoring and Evaluation

The monitoring and evaluation framework within Washington HB 1170, while not extensively detailed, includes a specific provision aimed at ensuring the ongoing effectiveness of AI detection tools. The legislation mandates that a covered provider shall collect user feedback related to the efficacy of its artificial intelligence detection tool. This feedback mechanism is crucial for identifying shortcomings, usability issues, or areas where the tool might be failing to accurately identify AI-generated or modified content. By actively soliciting input from the very users who rely on these tools, the bill establishes a direct channel for real-world performance assessment, moving beyond theoretical compliance to practical utility.

Furthermore, the bill requires covered providers to incorporate relevant feedback into any attempt to improve the efficacy of their detection tools. This provision underscores a commitment to continuous improvement and adaptability, recognizing that AI technology and the methods for detecting AI-generated content are rapidly evolving. Rather than imposing a static standard, the legislation encourages an iterative process where tools are refined based on practical experience and user needs. This dynamic approach to monitoring and evaluation is vital in a fast-paced technological landscape, ensuring that the transparency mechanisms remain effective against increasingly sophisticated AI systems. While broader governmental monitoring of overall compliance is implied through the Attorney General's enforcement powers, this specific requirement for user feedback on detection tools highlights a targeted mechanism for evaluating a core component of the bill's transparency objectives.

Penalties, Liability, and Appeals

Washington HB 1170 establishes a clear framework for penalties and enforcement by integrating its provisions within the existing Consumer Protection Act (Chapter 19.86 RCW). This means that any violation of the requirements set forth in HB 1170, such as failing to embed provenance data, provide AI detection tools, or ensure proper disclosures, can be considered an unfair or deceptive act or practice in commerce. Consequently, the Attorney General is empowered to bring actions against covered providers or other entities found in non-compliance. The remedies and penalties available under the Consumer Protection Act typically include injunctive relief, civil penalties, and the recovery of attorney's fees and costs. This approach provides a robust enforcement mechanism without necessitating the creation of new, AI-specific penalty structures, leveraging established legal avenues for consumer protection.

A significant aspect of the liability framework is the explicit declaration that any waivers of the provisions within HB 1170 are void. This ensures that entities cannot contractually exempt themselves from the transparency requirements, reinforcing the mandatory nature of the legislation and preventing attempts to circumvent its intent. While the bill does not detail a specific appeals process for enforcement actions, any actions brought by the Attorney General would follow the standard judicial procedures for cases under the Consumer Protection Act, allowing for due process and the opportunity for defendants to present their case in court. This integration into existing legal frameworks provides a familiar and established pathway for addressing non-compliance, ensuring that there are clear consequences for failing to adhere to the AI content transparency mandates.

Relationship to Other Instruments

Washington HB 1170 operates within a broader landscape of emerging AI regulations and voluntary industry standards, drawing parallels and building upon existing efforts. Notably, the bill's measured approach to AI content transparency is described as similar to California's Senate Bill 942 (SB 942), passed in 2024. This similarity suggests a growing trend in state-level legislation within the United States to address the challenges posed by AI-generated content, particularly deepfakes and misinformation. The alignment with California's law indicates an effort to create a consistent regulatory environment for companies operating nationally, potentially reducing the burden of disparate state-specific compliance requirements for AI providers. Such harmonization, even at a state level, can contribute to a more predictable regulatory landscape for the technology industry.

Furthermore, HB 1170 is designed to complement and build upon voluntary provenance standards that major technology companies are actively developing. Specifically, it references the Coalition for Content Provenance and Authenticity (C2PA), an industry-led initiative focused on developing technical specifications for content provenance and authenticity. By incorporating requirements that can be met through common standards like watermarking or metadata, the bill implicitly supports and encourages the adoption of such industry-wide technical solutions. This approach signifies a legislative strategy that seeks to reinforce and formalize best practices emerging from the private sector, rather than imposing entirely novel or conflicting technical mandates. This collaborative stance—acknowledging and integrating with existing and developing industry standards—aims to foster effective and widely adoptable solutions for AI content transparency.

International Alignment

While Washington HB 1170 primarily focuses on domestic regulation within the State of Washington, its underlying principles and technical requirements for AI content transparency resonate with broader international discussions and emerging regulatory trends. The emphasis on provenance data, AI detection tools, and clear disclosures for AI-generated content aligns with global efforts to combat misinformation, deepfakes, and the erosion of trust in digital media. Many international bodies and national governments worldwide are grappling with similar challenges posed by generative AI, leading to a growing consensus on the need for greater transparency and accountability in the development and deployment of AI systems.

Although the bill does not explicitly detail international partnerships or mutual recognition agreements, its alignment with voluntary industry standards like those proposed by the Coalition for Content Provenance and Authenticity (C2PA) suggests an indirect form of international alignment. C2PA is a cross-industry initiative with global participation, aiming to develop open technical standards for content provenance. By allowing compliance through such widely recognized standards, HB 1170 implicitly facilitates interoperability and consistency with international best practices. This approach helps ensure that AI systems and content originating from or distributed within Washington State can adhere to transparency norms that are increasingly becoming global expectations, contributing to a more trustworthy global digital ecosystem without imposing specific cross-border regulatory mandates.

Implementation Timeline

Sources and References