Takeaways from IAPP P.S.R. + AI Governance Global 2026: Desire for AI operational standards, shifting lens for analyzing third-party AI services, and continued focus on state privacy laws

The International Association of Privacy Professionals (IAPP) Privacy. Security. Risk. + AI Governance Global 2026 event was held October 6-9, 2026, in Seattle. Former U.S. presidential candidate and entrepreneur Andrew Yang opened the conference by describing a $2,500 offer to license his book for AI training. He contrasted it with the experience of most Americans, who have been paid nothing when their data may have been collected and used to train large language models. He suggested that disparity may help explain why many people are wary of AI.

The theme of that trust gap ran through the sessions we attended, from how organizations can credibly assess AI and how AI enters organizations through vendors and employees, to how Washington state’s attorney general is approaching what constitutes consent for data processing. Yet despite the open questions about earning consumer trust in AI, it was evident that many at the conference remain keen to build and deploy AI.

Below are a few key takeaways from sessions touching on this topic:

1. AI assurance: Strong desire for actionable standards and approach for assessing and mitigating AI risks but no clear agreed criteria or approach yet

Debate over whether 'AI assurance' is a field: Several panelists from standards- and credential-setting or similar organizations espoused the view that “AI assurance” (i.e., a field of professionals, likely third parties, with the credentials, background, and skills to evaluate, assess, measure, and opine as to whether AI systems meet applicable standards) is necessary and either already here or nearly so. They pointed to California laws that contemplate independent assessors in at least some circumstances and to voluntary commitments by frontier AI developers. Other panelists and attendees expressed skepticism for a variety of reasons: (1) Until there is an agreed set of standards for a given industry to measure against, there is no purpose for AI assurance. (2) There is, to date, no real difference between the skillsets and knowledge of in-house teams and so-called AI assurance professionals that would make it worth the money to pay for third-party AI assurance. (3) Absent a legal mandate (e.g., laws specifically regulating use of a frontier model), organizations are unlikely to establish a budget or internal cost center to pay for third-party AI assurance. (4) The many assessment or audit functions that companies already have (internal audit, third-party accounting, third-party cybersecurity) can address AI as part of their existing assessment or audit regimes.

What standards apply? Setting aside this debate, even those advocating for a profession of AI assurance agreed that at this time, there is not yet consensus on what an AI assessment should measure and against which standard. Few laws mandate AI assessments, those that do have limited applicability beyond frontier models, and even for frontier models the laws do not set out a detailed risk framework. And the current voluntary frameworks do not fill that gap on their own. ISO/IEC 42001 is a management-system standard: Like ISO 27001 for information security, it certifies that an organization has processes in place to govern AI (policies, roles, risk assessment, and monitoring), not that any particular AI system is accurate, fair, or safe. The NIST AI Risk Management Framework is widely referenced, but it does not prescribe pass/fail criteria.

What is the unit of measurement? At least one panelist recommended treating the use case rather than the tool or system as the unit of assessment because the same tool can present different risk profiles in different use cases.

Questions to consider before undertaking an AI assurance assessment or audit: When a vendor offers an “AI audit” or certification, or when you are considering commissioning one, consider asking:

• Is it testing, a risk assessment, or an audit against fixed criteria?
• What are the criteria?
• Who set the criteria?
• Who is the reviewer independent of?
• What is your organization’s basis for doing this? Does an applicable law require it, or is a business partner or internal stakeholder pushing for it? Who will pay for it? Does your organization have the internal expertise to do the same work, and would that be appropriate for the use case?

Also keep in mind that an assurance report on a vendor’s model does not necessarily cover how you deploy it. Panelists said it is hard to separate a model from its data and the system it sits in. Ask vendors for the evidence they already have and plan your own use-case review of reliability, privacy, and confidentiality in your organization’s environment.

2. AI risk from vendor software often enters outside the moment of procurement

Several sessions addressed third-party AI vendor risk management, and the large audience for the topic suggested it is becoming a business priority. The push to adopt AI at speed and by all business stakeholders in an organization can risk leaving organizations without a clear picture of which AI tools and vendors are actually in use. One panelist made the point by asking the room how many AI vendors the audience’s organizations have and then citing a 2025 survey in which 69% of security leaders suspected use of banned public generative AI in their organizations.

Speakers identified other drivers of this risk. Traditional vendor due diligence and compliance frameworks are unlikely to fully account for the particular characteristics of AI, including its dynamic nature and the complexity of the AI value chain. Growing, but still limited, employee awareness of AI risks and best practices leads to employees using tools (including privately paid subscription AI tools) that have not undergone supplier review process. The list goes on. For example, a key issue is whether the vendor contractually has the ability to train AI on their customer’s data. Whether customers push back on this varies.

AI supplier contracting “best practices” are likely to evolve quickly. Suppliers and customers are becoming more sophisticated, and legal positions and internal policies will increasingly reflect lessons from deployment, especially where things have gone wrong. Regulation, such as the many state-level AI laws in the United States and the EU AI Act, is likely to be another catalyst for increased scrutiny of supplier relationships. The result will be a much more mature, harder-to-negotiate approach to AI supplier risk that accounts for the different players in the AI value chain.

3. Washington state AG: A comprehensive privacy law is the goal but not in 2027

In August 2026, Attorney General Nick Brown’s "Data Privacy Report" listed legislative asks without endorsing a comprehensive law on data privacy (see our summary here). At the conference, he said he thinks Washington state needs a comprehensive privacy law to join the state’s groundbreaking health privacy law, the My Health, My Data Act. However, he does not expect one in the January 2027 session, and he described his office’s recent privacy report as an effort to restart that conversation. The discussion raised the previous efforts of Washington state to enact privacy legislation as a model that has been adopted elsewhere but has not resulted in comprehensive privacy laws in the state.

Consent: Brown in his report advocates for meaningful, informed consent. At IAPP, Brown said he expects exceptions to be “on the table” and wants tech stakeholders involved in those discussions. That gives businesses a reason to engage before any legislative text is proposed.

The survey: Brown noted that people want the convenience of sharing data across services, even when they tell pollsters they don’t want their data shared. That potential tension is useful context for reading the report’s survey figures.

Enforcement in the meantime: Brown described his job as enforcing what statutes the legislature has enacted. Expect consent flows, deceptive design, overcollection, and sensitive data to be scrutinized under the Washington Consumer Protection Act. He also said the current law limits his enforcement tools, which tracks the report’s recommendation that the legislature revisit remedies and signals where his office will push next.