Independent Verification as a Practical Accountability Model for Advanced AI
California has taken a significant step towards answering an important AI governance question. Who should evaluate whether an AI developer’s safety and risk claims can be trusted?
On September 9, Governor Gavin Newsom signed California Senate Bill 813, which requires California officials to establish requirements for designating and overseeing Independent Verification Organizations, or IVOs.
This action builds a state framework for qualified third parties to assess the risks posed by AI systems and models and identifies the metrics and methodologies supporting their conclusions.
The legislation is particularly significant in the context of California’s frontier AI framework. Large frontier developers are already required to publish frameworks describing how they assess and manage potentially catastrophic risks. But much of the underlying evaluation is still performed by the companies developing the systems.
SB 813 points toward independent scrutiny of the controls, processes, and claims on which public trust increasingly depends, the next layer of AI governance.
But identifying qualified independent evaluators is only the beginning. The larger policy challenge is how to build a durable accountability system around their work. Independence matters, but so do clear standards, meaningful review, corrective action, continuing oversight, and consequences when either a company or its verifier falls short.
Government agencies, however, may not always have the specialized expertise, capacity, or flexibility to conduct every technical assessment themselves. AI technology and evaluation methods also evolve quickly, making it difficult for legislation to prescribe a single test or methodology that will remain effective over time.
IVOs offer a potential bridge. Qualified independent organizations can apply specialized expertise, examine technical and organizational controls, test claims, document findings, and identify gaps that may not be apparent from public disclosures alone. Government can establish the requirements for independence and performance while retaining oversight and enforcement authority.
SB 813 would lay the groundwork for this model by requiring California’s Government Operations Agency to develop criteria for IVO designation. Those criteria would address technical competence, assessment methodologies, conflicts of interest, and independence from the organizations being evaluated. The agency would also establish procedures for suspending or terminating an IVO’s designation for issues such as material misrepresentations, inadequate documentation, compromised independence, or conduct that calls its competence or integrity into question.
The bill would not require AI developers to use an IVO or undergo an audit. Instead, it would begin building the infrastructure necessary for policymakers to rely on independent AI assessments in future laws or programs.
At the federal level, the proposed FRONTIER Act (introduced by Reps. Jay Obernolte (R-CA) and Lori Trahan (D-MA)) would establish a more extensive framework for licensing and overseeing IVOs that assess certain large frontier AI developers. The proposal addresses not only IVO qualifications and independence, but also assessment requirements, corrective action, continuing oversight, and circumstances that could result in the loss of an IVO’s license.
These proposals differ in scope, but they both indicate that when the risks are consequential and the technology is highly complex, company self-assessment cannot be the only source of assurance.
The principle can also extend beyond a single class of AI model. Frontier, open-weight, and other advanced AI systems may require different assessment methods based on their capabilities, distribution, downstream uses, and risk profiles. An effective independent verification framework should preserve that flexibility while requiring claims about safeguards and risk mitigation to be supported by evidence that qualified outside parties can examine.
BBB National Programs has spent decades operating independent accountability mechanisms in areas where trust depends on more than an organization’s own assurances. Through government-recognized programs supporting children’s privacy and cross-border data flows, we have seen that independent review is most effective when it operates as part of a broader system.
BBB National Programs operates the nation’s first FTC-approved COPPA Safe Harbor program. The COPPA safe harbor model demonstrates how government-approved guidelines can be translated into meaningful review, ongoing monitoring and compliance, and enforcement—not simply a one-time certification.
Cross-border privacy systems use related accountability structures. BBB National Programs serves as the longest-running U.S. Independent Recourse Mechanism under the Data Privacy Framework, helping participating organizations meet their obligations while providing independent dispute resolution for individuals. We also serve as an approved Accountability Agent under the Global Cross-Border Privacy Rules system, independently certifying organizations against established privacy requirements and monitoring their continued compliance.
In both frameworks, independent accountability operates alongside government administration and enforcement.
These programs address different risks and legal obligations than advanced AI, and none offers a perfect template for IVOs. But they demonstrate that government recognition of an independent organization is not itself accountability.
Accountability comes from what the framework requires that organization and participating companies to do over time. Who establishes the standards? What evidence must be reviewed? How are deficiencies corrected? What happens if a participant refuses to act? How is the independent organization itself monitored? Where can concerns be escalated? And what consequences follow when the system does not perform as intended?
Those operational questions will determine whether IVOs become trusted accountability partners or simply another category of third-party auditor.
The next task is to ensure that independent verification becomes more than a technical checkpoint. A credible IVO framework must connect assessment to correction, continuing oversight, and consequences. It must hold AI developers accountable for responding to identified risks while also holding verifiers accountable for the quality and independence of their work.
Privacy and other co-regulatory systems show that government does not need to conduct every assessment itself to preserve meaningful public oversight. It can establish clear expectations, recognize qualified accountability partners, require continuing performance, and retain enforcement authority.
If California builds those elements into its IVO framework, SB 813 could do more than create a new class of AI auditors. It could provide a foundation for independent AI accountability that is credible, adaptable, and capable of earning public trust.
On September 9, Governor Gavin Newsom signed California Senate Bill 813, which requires California officials to establish requirements for designating and overseeing Independent Verification Organizations, or IVOs.
This action builds a state framework for qualified third parties to assess the risks posed by AI systems and models and identifies the metrics and methodologies supporting their conclusions.
The legislation is particularly significant in the context of California’s frontier AI framework. Large frontier developers are already required to publish frameworks describing how they assess and manage potentially catastrophic risks. But much of the underlying evaluation is still performed by the companies developing the systems.
SB 813 points toward independent scrutiny of the controls, processes, and claims on which public trust increasingly depends, the next layer of AI governance.
But identifying qualified independent evaluators is only the beginning. The larger policy challenge is how to build a durable accountability system around their work. Independence matters, but so do clear standards, meaningful review, corrective action, continuing oversight, and consequences when either a company or its verifier falls short.
From Company Assurances to Independent Evaluation
Frontier AI models raise especially urgent governance questions because of their scale, capabilities, and potential risks. Policymakers must be able to assess whether developers are identifying those risks, implementing appropriate safeguards, documenting their decisions, and responding effectively when deficiencies or incidents emerge.Government agencies, however, may not always have the specialized expertise, capacity, or flexibility to conduct every technical assessment themselves. AI technology and evaluation methods also evolve quickly, making it difficult for legislation to prescribe a single test or methodology that will remain effective over time.
IVOs offer a potential bridge. Qualified independent organizations can apply specialized expertise, examine technical and organizational controls, test claims, document findings, and identify gaps that may not be apparent from public disclosures alone. Government can establish the requirements for independence and performance while retaining oversight and enforcement authority.
SB 813 would lay the groundwork for this model by requiring California’s Government Operations Agency to develop criteria for IVO designation. Those criteria would address technical competence, assessment methodologies, conflicts of interest, and independence from the organizations being evaluated. The agency would also establish procedures for suspending or terminating an IVO’s designation for issues such as material misrepresentations, inadequate documentation, compromised independence, or conduct that calls its competence or integrity into question.
The bill would not require AI developers to use an IVO or undergo an audit. Instead, it would begin building the infrastructure necessary for policymakers to rely on independent AI assessments in future laws or programs.
A Broader Policy Movement
California is not alone in considering how independent organizations can support public oversight of advanced AI.At the federal level, the proposed FRONTIER Act (introduced by Reps. Jay Obernolte (R-CA) and Lori Trahan (D-MA)) would establish a more extensive framework for licensing and overseeing IVOs that assess certain large frontier AI developers. The proposal addresses not only IVO qualifications and independence, but also assessment requirements, corrective action, continuing oversight, and circumstances that could result in the loss of an IVO’s license.
These proposals differ in scope, but they both indicate that when the risks are consequential and the technology is highly complex, company self-assessment cannot be the only source of assurance.
The principle can also extend beyond a single class of AI model. Frontier, open-weight, and other advanced AI systems may require different assessment methods based on their capabilities, distribution, downstream uses, and risk profiles. An effective independent verification framework should preserve that flexibility while requiring claims about safeguards and risk mitigation to be supported by evidence that qualified outside parties can examine.
What Privacy Accountability Models Can Teach AI Policymakers
The governance challenge may be new in the context of frontier AI, but the underlying accountability architecture is not entirely new.BBB National Programs has spent decades operating independent accountability mechanisms in areas where trust depends on more than an organization’s own assurances. Through government-recognized programs supporting children’s privacy and cross-border data flows, we have seen that independent review is most effective when it operates as part of a broader system.
BBB National Programs operates the nation’s first FTC-approved COPPA Safe Harbor program. The COPPA safe harbor model demonstrates how government-approved guidelines can be translated into meaningful review, ongoing monitoring and compliance, and enforcement—not simply a one-time certification.
Cross-border privacy systems use related accountability structures. BBB National Programs serves as the longest-running U.S. Independent Recourse Mechanism under the Data Privacy Framework, helping participating organizations meet their obligations while providing independent dispute resolution for individuals. We also serve as an approved Accountability Agent under the Global Cross-Border Privacy Rules system, independently certifying organizations against established privacy requirements and monitoring their continued compliance.
In both frameworks, independent accountability operates alongside government administration and enforcement.
These programs address different risks and legal obligations than advanced AI, and none offers a perfect template for IVOs. But they demonstrate that government recognition of an independent organization is not itself accountability.
Accountability comes from what the framework requires that organization and participating companies to do over time. Who establishes the standards? What evidence must be reviewed? How are deficiencies corrected? What happens if a participant refuses to act? How is the independent organization itself monitored? Where can concerns be escalated? And what consequences follow when the system does not perform as intended?
Those operational questions will determine whether IVOs become trusted accountability partners or simply another category of third-party auditor.
The Next Layer of AI Governance
Consequential claims about advanced systems should not simply be taken on faith.The next task is to ensure that independent verification becomes more than a technical checkpoint. A credible IVO framework must connect assessment to correction, continuing oversight, and consequences. It must hold AI developers accountable for responding to identified risks while also holding verifiers accountable for the quality and independence of their work.
Privacy and other co-regulatory systems show that government does not need to conduct every assessment itself to preserve meaningful public oversight. It can establish clear expectations, recognize qualified accountability partners, require continuing performance, and retain enforcement authority.
If California builds those elements into its IVO framework, SB 813 could do more than create a new class of AI auditors. It could provide a foundation for independent AI accountability that is credible, adaptable, and capable of earning public trust.