Summit Law Blog

Remember the Washington State AI Task Force? Well, since it published the Inaugural Report, it has been busy–busy proposing and reviewing at least eight policy recommendations so far regarding the development, deployment, and use of AI in Washington State.[1]

One recommendation in particular, Recommendation #2 on Transparency in Training Data (“Recommendation”), was fully approved on September 25^th. Let’s discuss this Recommendation, how HB 1168 ties in, and what California has to do with all of this.

What is Recommendation #2 on Transparency in Training Data?

The Joint Ethical AI/AI Governance and Consumer Protection & Privacy Subcommittee (“Subcommittee”) proposed this Recommendation. Initially discussed during the August 21^st meeting, the AI Task Force conditionally approved the Recommendation, subsequently amended it, discussed it, and fully approved it during the September 25^th meeting.

Forming the basis of the Recommendation, the Subcommittee outlined its findings, explaining what transparency means, why it is important, and how it relates to training data. Specifically, the Subcommittee explained that transparency:

• means “making the processes and decisions behind AI systems clear and understandable to the public, users, and regulators” while also “giving notice to users and consumers when they are engaged with or impacted by AI where appropriate.”^[2]
• is important, both for the public and for private companies, because it builds trust, ensures accountability, mitigates potential harms and bias, and promotes ethical practices.^[3]
• is important (as it relates to training data) because training data forms the basis of AI systems, so the quality, quantity, and diversity of training data “is essential to producing reliable results.”^[4]

However, acquiring more data raises privacy, intellectual property, and ethical considerations.^[5] Based on these findings, the Subcommittee proposed two specific recommendations.[6]

Recommendation 2.1:

The Subcommittee recommended that the Washington State Legislature require AI developers make public disclosures that describe “the provenance,[7] quality, quantity, and diversity of datasets used for training AI models.”^[8] This disclosure requires details such as:

• the source of the data and the method of acquisition;
• clear metrics on the quantity and types of data;
• the processes used to prepare and annotate data prior to processing; and
• assessment of data representation across relevant factors such as demographics, content types, and language.^[9]

Recommendation 2.2:

Additionally, the Subcommittee recommended that the Legislature require AI developers to make public disclosures, explaining “how training data is processed to mitigate errors and biases during AI model development.”^[10] This disclosure requires information on:

• how data is assessed for potential bias before training, and strategies for mitigation;
• how sensitive personal data is identified and processed to prevent discrimination or privacy breaches; and
• the pipeline for model development, including how processing of different model versions is distinguished and managed.^[11]

For both recommendations, trade secret and proprietary information is exempt from disclosure.^[12]

By now, you might be thinking, this sounds familiar – didn’t the Washington State Legislature introduce similar legislation during the 2025 legislative session? Yes, good memory. 

What is HB 1168?

Washington State’s House Bill 1168, An Act Relating to Increasing Transparency in Artificial Intelligence (“HB 1168”), was a legislative bill, proposed during the 2025 Regular Session by Representatives Shavers, Taylor, Ryu, and Fosse.^[13] If passed, HB 1168 would have required AI developers to post documentation regarding the data used to train AI systems or services.[14]

More specifically, HB 1168^[15] would have required developers to post documentation regarding the data used to train generative AI systems or services (released or substantially modified on or after January 1, 2022) before such systems or services were made publicly available. The documentation would have been required to summarize the training datasets and include the following information:

• Source or owner of data;
• Dataset’s relevance to the AI system’s or service’s purpose;
• Number and description of types of data points;
• Whether data was public, purchased, or licensed;
• Presence of personal or aggregate consumer information;
• Any modifications to the datasets;
• Date of initial training or last major update;
• Use of synthetic data generation.^[16]

Exemptions would have included generative AI systems and services developed exclusively for security, aviation operations, or national defense purposes, and made available only to a federal entity. Failure to comply under HB 1168 would have resulted in a penalty of $5,000 per day, enforceable through a civil action brought by the Attorney General; although, the Attorney General would have been required to provide a 45-day cure period if they had determined that a cure is possible.

However, HB 1168 did not pass^[17] and this is where California gets bragging rights.

What’s California Got to Do with This?

As subtly noted in the Staff Summary of HB 1168’s Public Testimony,[18] HB 1168 was based on California’s Assembly Bill No. 2013 on Generative Artificial Intelligence; Training Data Transparency (“AB 2013”).^[19] Although nearly identical, there are some key differences between the two bills:

• HB 1168 removed the requirement of identifying “whether the datasets include any data protected by copyright, trademark, or patent.”[20]
• HB 1168 removed the requirement of identifying “the time period during which the data in the datasets were collected, including a notice if the data collection is ongoing.”[21]
• HB 1168 added a requirement to identify the “date of the last significant update to the datasets during the development” of the AI system or service.[22]
• And finally, unlike HB 1168, AB 2013 was passed and signed into law.

There are some advantages to Washington following California’s lead. First, as noted in the Staff Summary, California’s full-time legislators have more time than Washington’s part-time legislators to work on bills.[23] Second, Washington can observe industry and court responses to California’s law and use those responses to improve its version of the same law. Ultimately, if and when Washington passes its version of this law, it will hopefully not be viewed as a political reaction to the rise of AI, but rather as an intentionally thought-out approach.

Tying it All Together

To wrap up this discussion, here are three key takeaways to keep in mind.

First, if you are developing, or planning to develop, AI products, services, or systems in Washington State, begin documenting your training datasets now, taking into account the specific information required (see above). That way, you will be better prepared to comply with HB 1168 if it passes. If you plan to expand into other jurisdictions, watch for similar or even more stringent requirements, such as those passed in California. As always, give us a call if you have any compliance questions or issues.

Second, as Washington approaches its 2026 Legislative Session, we’ll be watching HB 1168 to see whether it is reintroduced and what changes lawmakers propose. Specifically, we will examine California’s experience with AB 2013 and how newer California AI laws (such as SB 53,[24] which requires developers of large frontier models to publish safety plans and report critical safety incidents) affect HB 1168’s direction. Additionally, with the addition of the AI Task Force’s Recommendation, we will look to see whether the following anticipated changes are addressed:

• A new disclosure requirement for assessing data representation across relevant factors, such as demographics, content types, and language.
• A new disclosure requirement related to “how training data is processed to mitigate errors and biases during AI model development.”^[25]
• An explicit exemption protecting trade secrets and proprietary information from mandated disclosure.
• A prohibition on private right of action.

Third, even if HB 1168 does not pass, these ongoing efforts reflect consumers’ growing expectations of transparency and accountability from AI developers and organizations. Even if not legally required, AI developers and organizations should proactively embed transparency into their AI governance frameworks to demonstrate leadership in responsible AI development and, more importantly, build trust. Washington’s AI industry is among the leaders in AI and should continue to lead the way by modeling transparency, accountability, and ethical innovation.

The next report will be released on December 1, 2025. We will continue to report on additional recommendations and updates as they become available. As a reminder, there is still time to participate in the Task Force. If you would like to participate, you can submit written comments at any time by emailing AI@atg.wa.gov. You can also provide public comments during a meeting by emailing AI@atg.wa.gov at least 24 hours prior to the meeting.

For any questions, please reach out to our Tech Transactions team!

---

[1] E-mail from ATG MI ADM AITF, to author (Aug. 14, 2025, 3:16 PST) (on file with author); e-mail from ATG MI ADM AITF, to author (Sept. 17, 2025, 10:59 PST) (on file with author).

[2] Joint Ethical AI/AI Governance and Consumer Prot. & Priv. Subcomm., Wash. AI Task Force, Recommendation #2 Findings § 1 (2025) (on file with author).

[3] Id. Findings §§ 1, 10.

[4] Id. Findings §§ 3–4.

[5] Id. Findings § 9.

[6] Confusingly, the Subcommittee lists two recommendations, numbered 1 and 2, under Recommendation #2. For clarity, I’ll refer to them here as Recommendations 2.1 and 2.2.

[7] “Data provenance is the historical record of data that details data’s origins by capturing its metadata as it moves through various processes and transformations.” Tim Mucci, What is Data Provenance?, IBM, https://www.ibm.com/think/topics/data-provenance (last visited Nov. 13, 2025).

[8] Joint Ethical AI/AI Governance and Consumer Prot. & Priv. Subcomm., Wash. AI Task Force, supra note 2, Recommendations § 1.

[9] Joint Ethical AI/AI Governance and Consumer Prot. & Priv. Subcomm., Wash. AI Task Force, supra note 2, Recommendations § 1.

[10] Joint Ethical AI/AI Governance and Consumer Prot. & Priv. Subcomm., Wash. AI Task Force, supra note 2, Recommendations § 2.

[11] Joint Ethical AI/AI Governance and Consumer Prot. & Priv. Subcomm., Wash. AI Task Force, supra note 2, Recommendations § 2.

[12] Joint Ethical AI/AI Governance and Consumer Prot. & Priv. Subcomm., Wash. AI Task Force, supra note 2, Recommendations §§ 1–2.

[13] Wash. State House Comm. on Tech., Econ. Dev., & Veterans, House Bill Report HB 1168, 2025 Reg. Sess. at 1 (2025), https://lawfilesext.leg.wa.gov/biennium/2025-26/Pdf/Bill%20Reports/House/1168%20HBR%20TEDV%2025.pdf?q=20250130021627.

[14] H.B. 1168, 69th Leg., 2025 Reg. Sess. (Wash. 2025), https://lawfilesext.leg.wa.gov/biennium/2025-26/Pdf/Bills/House%20Bills/1168-S.pdf?q=20251113154701.

[15] For the purposes of this blog, I will be discussing Substitute House Bill 1168.

[16] H.B. 1168, 69th Leg., 2025 Reg. Sess. (Wash. 2025), https://lawfilesext.leg.wa.gov/biennium/2025-26/Pdf/Bills/House%20Bills/1168-S.pdf?q=20251113154701.

[17] HB 1168 – 2025-26, Wash. State Legis., https://app.leg.wa.gov/billsummary?BillNumber=1168&Year=2025&Initiative=False (last visited Nov. 13, 2025).

[18] Wash. State House Comm. on Tech., Econ. Dev., & Veterans, supra note 13, at 5.

[19] A.B. 2013, 2023–2024 Reg. Sess. (Cal. 2024), https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202320240AB2013.

[20] Id.

[21] Id.

[22] H.B. 1168, 69th Leg., 2025 Reg. Sess. (Wash. 2025), https://lawfilesext.leg.wa.gov/biennium/2025-26/Pdf/Bills/House%20Bills/1168-S.pdf?q=20251113154701.

[23] Wash. State House Comm. on Tech., Econ. Dev., & Veterans, supra note 13, at 5.

[24] S.B. 53, 2025–2026 Reg. Sess. (Cal. 2025), https://legiscan.com/CA/text/SB53/id/3270002.

[25] Joint Ethical AI/AI Governance and Consumer Prot. & Priv. Subcomm., Wash. AI Task Force, supra note 2, Recommendations § 2.