Mandate third-party audits for major AI systems

ai-governance ai-policy ai-regulation policy regulations transparency

Share this with your representatives, journalists and friends

Key insights

Both supportive and critical quotes emphasize standards and qualified auditors; they differ mainly on whether those conditions are developed enough for a mandate now.
Many supportive quotes link audits to transparency reports, incident reporting, certification, and pre- and post-deployment testing, rather than treating audits as a standalone fix.
Several proponents explicitly scope audit requirements to the largest, frontier, or high-risk systems, while several critics focus their cost concerns on smaller deployers and businesses.
Even some skeptical or cautious quotes still point toward independent verification through voluntary audits, feasibility studies, or federally overseen organizations, suggesting disagreement over sequencing as well as principle.
In the sample, supportive quotes come from elected officials, regulators, AI companies, researchers, nonprofits, and international bodies, while critical quotes are concentrated among trade associations and business advocates.

Arguments for

Trust through independence

Proponents say trust cannot rest on self-assessment, developers' goodwill, or testers with close ties to the labs they evaluate. Independent third-party review is presented as practical verifiability that builds objective accountability and public trust.

It is really important, given the stakes, that people who develop AI and those who test it don’t have relationships with each other.

Yoshua Bengio · Sep 1, 2026 · full quote
The current regime is definitely not optimal. While third-party evaluators try hard to be unbiased, a lab can decide to stop working with them any day for any reason whatsoever. We really need more stringent regulations to improve the incentives for evaluations.

Marius Hobbhahn · Sep 1, 2026 · full quote
Trust cannot rest on self-assessment or voluntary disclosure alone. It requires the ability to independently assess systems through auditability, traceability, and validation mechanisms across the AI lifecycle – drawing on established practices from cybersecurity, safety engineer…

Daphné Barbotte · May 14, 2026 · full quote
Our Witness Tried benchmark shows that detection currently works least well for those most at risk, particularly in the global majority. Both must rest on independent third-party oversight. Trust verified by third parties, not asserted by the developers who profit from them.

Sam Gregory · Jul 7, 2026 · full quote
Self-attestation is not governance. The Chicago Principles describe what assurance must be to deserve the trust markets are now asked to place in it.

AIQA Global · May 28, 2026 · full quote

Safety guardrail for high-risk

Supporters argue that the stakes are high enough to require mandatory independent testing before release, regular audits afterward, and related oversight such as incident reporting. In this view, audits are foundational safeguards for frontier systems whose failures could endanger public safety or exceed acceptable risk thresholds.

In the face of a stunning failure, OpenAI appears to be taking steps that prioritize the performance and profit of its A.I. models with the knowledge that those changes could be detrimental to public safety. This demonstrates the need for vigorous, mandatory independent auditing …

Richard Blumenthal · Sep 9, 2026 · full quote
We need regular mandatory independent safety testing and oversight, mandatory disclosure of security incidents, and international cooperation to keep people safe from absolute disaster.

Greg Casar · Jul 23, 2026 · full quote
Secondly, the Government must mandate independent statutory testing for AI before its release, and a yellow card scheme that would be a national register where safety incidents, near misses and harms caused by AI were reported, tracked and acted on exactly as we do for adverse dr…

Iqbal Mohamed · Jul 8, 2026 · full quote
Our Advanced AI Framework proposes mandatory independent safety testing for frontier models, transparency requirements, and government authority to block or recall dangerous AI deployments.

Anthropic · Jun 12, 2026 · full quote
Even the companies developing them seem unable to reliably predict or control the systems’ behavior. Against that backdrop, requirements like transparency, testing and auditing should probably be understood as foundational safeguards rather than the upper limit of accountability.

Jonathan Iwry · May 22, 2026 · full quote

Governance infrastructure

Many supporters treat audits as part of a broader infrastructure of standards, certification, oversight, and reporting, similar to other domains where trust depends on institutions rather than simplicity. They argue that mandates can help create durable markets and institutions of independent verifiers that governments and the public can rely on.

If we place our trust in these products or machines, it is not because they are simple, transparent or harmless. It is because, over decades, we have built a genuine infrastructure of trust around them: laws, regulations, oversight authorities, certification procedures, audits, s…

Olivier Sibony · Jul 22, 2026 · full quote
By establishing robust safety standards for developers, requiring independent assurance of those standards, and ensuring transparency into frontier AI development, the federal government can give the public meaningful protections while preserving America’s leadership in AI.

Brad Carson · Aug 10, 2026 · full quote
Independent third-party AI testing and assurance providers play an important role in translating AI safety research into repeatable evaluation methods, objective testing reports, certification schemes, and practical risk management support for industry.

The 2026 Singapore Consensus on Global AI Safety Research Priorities · Jul 2026 · full quote
What will work is a competitive market of independent verifiers who are accountable for real-world outcomes and who the government can actually count on.

Andrew Freedman · Jul 23, 2026 · full quote
Ensuring that external evaluators meet appropriate independence and qualification requirements is critical for the credibility and effectiveness of this evaluation framework.

European AI Office · Jun 12, 2026 · full quote

Targeted rules for major developers

Some proponents stress that audit mandates can be aimed at the largest or most advanced developers rather than applied as one-size-fits-all rules. They argue that pairing independent assessments with published safety frameworks and transparency can protect the public while preserving room for smaller innovators and continued leadership.

By tying coverage to substantial AI research and development investments and requiring transparency and independent assessments from those developers, the bill avoids one-size-fits-all mandates and allows smaller innovators to spend their time building new technologies instead of…

Scott Franklin · Jul 23, 2026 · full quote
Artificial intelligence is advancing rapidly, and Congress must ensure our regulatory framework keeps pace without undermining American innovation. The FRONTIER Act focuses oversight on the largest developers and most advanced models, requiring transparency, independent evaluatio…

Jay Obernolte · Jul 23, 2026 · full quote
Require the largest AI companies to publish and comply with safety frameworks, submit to third-party audits, and ensure whistleblowers are protected.

Lori Trahan · Jun 1, 2026 · full quote
Large frontier developers should annually retain an independent third party to audit compliance with frontier safety requirements, including implementation of the developer’s frontier AI framework, internal controls, and governance structures.

OpenAI · Jun 2, 2026 · full quote
Companies developing the most capable systems should meet clear requirements, including independent audits, incident reporting, strong security standards, and whistleblower protections.

Chris Lehane · Jul 15, 2026 · full quote

Arguments against

Immature audit ecosystem

Opponents argue a mandate is premature because there are no recognized frontier-model auditing standards, certified auditors, established methodologies, or enough qualified vendors. On that view, companies would face an unrealistic or highly subjective compliance obligation on an aggressive timeline.

Illinois SB 315’s mandatory third-party AI audit requirement creates an impossible compliance burden — there are no recognized auditing standards, certified auditors or established methodologies for frontier model safety audits, yet companies would face penalties up to $3 million…

Amy Bos · May 26, 2026 · full quote
SB 315 would require third-party audits of AI systems, but a sufficient market of qualified auditors does not yet exist. Companies could face significant compliance risk because the vendors needed to meet the bill are not widely available. In addition, there is no consensus on th…

Consumer Technology Association · May 22, 2026 · full quote
This bill contains a critically flawed mandatory third-party audit requirement that creates an impossible compliance obligation on an unrealistic timeline.

NetChoice · May 26, 2026 · full quote
We remain concerned that Illinois would effectively be requiring private actors to make highly subjective determinations requiring AI safety compliance without established national standards, certifications, or clear regulatory guardrails.

Ninia Linero · May 20, 2026 · full quote
SIIA raised particular concerns about applying traditional product liability frameworks to generative AI, regulating frontier models at the state level, and imposing premature audit and transparency requirements without established standards.

Software & Information Industry Association · Apr 13, 2026 · full quote

Compliance cost spillovers

Critics say repeated third-party audits, reporting, and disclosures may be manageable for large organizations but prohibitively expensive for small and mid-sized businesses. They also warn that costs borne by major developers could be passed through in higher fees to downstream users such as hospitals, banks, law firms, manufacturers, and other businesses.

While the Big Tech developers will easily absorb the audit costs or pass them on through their pricing to consumers, every Illinois business that uses American AI — every hospital, every community bank, every law firm, every manufacturer — will have to pay for that audit in highe…

Jeremy Kudon · May 29, 2026 · full quote
Independent AI audits currently cost tens of thousands of dollars per system, depending on complexity. For employers using multiple HR technology tools, the cost could easily reach six figures annually or more. Small and mid-sized employers, which make up the majority of Connecti…

Paul Amarone · Mar 10, 2026 · full quote
Repeated third-party audits, extensive reporting, and public disclosures may be manageable for large organizations, but for small and mid-sized employers — which make up most New York workplaces — these costs could be prohibitive.

Alexander Alonso · Jan 15, 2026 · full quote
The onerous audit and reporting requirements for both AI developers and AI deployers are concerning for several reasons. The bill requires significant and costly third-party audits to be conducted by both developers and deployers. For any business to comply, they will be forced t…

Chelsea Lemon · Jan 15, 2026 · full quote

Confidentiality risks

Some opponents say aggressive audit and data-sharing requirements could expose entrepreneurs' trade secrets and private records. They treat protection of confidential business information as a separate reason to resist current mandate designs.

There are some changes we’d like to work with the sponsors on, including the bill’s aggressive auditing regime and data-sharing requirements for developers that could risk entrepreneurs’ trade secrets and private records.

Patrick Hedger · Jun 4, 2026 · full quote

Middle ground

Phase-in approach

Some participants support outside evaluation in principle but suggest starting with voluntary sharing or voluntary audits for the most advanced models, overseen by an independent or federally overseen body. They argue binding mandates should follow only after the regime proves effective and robust or after governments study feasibility, standards, risks, and implementation challenges.

Frontier labs would initially share their models with the body voluntarily, up to 30 days before release, for safety testing that probes dangerous cyber, biological and deception capabilities.

Demis Hassabis · Jul 14, 2026 · full quote
Once the testing regime proves “effective and robust,” formalization “could quickly follow.” That means frontier models would be required to pass before they could be deployed in the U.S. market.

Demis Hassabis · Jul 14, 2026 · full quote
For the most advanced frontier AI models, an independent, federally overseen, industry-backed organization could set safety standards and verify independent, voluntary audits, similar to how other critical sectors are managed.

Kent Walker · Jun 25, 2026 · full quote
Rather than mandating audits before a mature verification ecosystem exists, legislators should consider an approach similar to legislation adopted in Virginia and Connecticut, which first directs state agencies to study the feasibility, standards, risks, and implementation challe…

Aden Hizkias · May 27, 2026 · full quote

Across the quotes, outside scrutiny of advanced AI systems is often described as valuable, but there is a clear divide over whether it should already be mandatory. Supporters frame mandatory third-party audits for major or frontier systems as a baseline safeguard and a way to build trustworthy governance infrastructure. Opponents and cautious voices argue that until standards, auditor capacity, timelines, and confidentiality protections are clearer, mandates may be premature or overly burdensome.

Cast your vote:

Results (227 votes):

Total (227 votes)

For 206 (91%) Abstain 0 (0%) Against 21 (9%)

For (201)

Raj Panjabi
Physician, global-health and biodefense expert; former White House Senior Director for Global Health Security and Biodefense.

I am this person! Add a sourced quote Report comment

votes For and says:

That requires independent evaluators to red-team AI models before deployment, not voluntary company disclosures.
AI Verified source (Oct 9, 2026)

added 1d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Maria Cantwell
U.S. Senator from Washington

I am this person! Add a sourced quote Report comment

votes For and says:

Frontier AI systems require constant oversight and vigilance at home and abroad. The United States should establish clear, transparent, enforceable safety standards, developed by federal experts at NIST in coordination with other relevant federal age...
more AI Verified source (Oct 7, 2026)

added 2d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
V. Ramgopal Rao
Indian nanoelectronics researcher, IEEE Fellow, and Vice-Chancellor of BITS Pilani.

I am this person! Add a sourced quote Report comment

votes For and says:

Independent audits could help, but there needed to be clarity over who selects auditors, what standards they use, how much access they receive and whether findings are made public. [...] Some of these voluntary commitments would eventually need statu...
more AI Verified source (Oct 1, 2026)

added 10d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Rob Reich
Political scientist and professor at Stanford

I am this person! Add a sourced quote Report comment

votes For and says:

Developers should not be the sole judges of the systems they build. It’s like asking students to grade their own homework. Independent evaluations, conducted by government or qualified third parties, should be a standard part of the AI ecosystem.
AI Verified source (Sep 22, 2026)

added 17d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Asad Baig
Technology-policy advocate, journalist, and founder of Media Matters for Democracy.

I am this person! Add a sourced quote Report comment

votes For and says:

A better approach would be risk-based. Low-risk applications need little intervention, while systems crossing defined thresholds in areas such as cyber capability, biological risk, weapons, critical infrastructure or autonomous action should face muc...
more AI Verified source (Sep 21, 2026)

added 19d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
BinaryBriefing
Daily technology news and reference publication covering AI, cybersecurity, technology policy, and emerging technology.

I am this person! Add a sourced quote Report comment

votes For and says:

It is an argument for treating incident discovery and independent review as part of the safety system itself. If advanced agents can generate vast volumes of activity quickly, companies need clear thresholds for what gets escalated, investigated, ind...
more AI Verified source (Sep 19, 2026)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Martin Novak
Founder of the Synthocracy Institute; author and independent researcher on AI, governance, institutional power, and decision systems.

I am this person! Add a sourced quote Report comment

votes For and says:

The governance direction is therefore already visible: developer evidence → independent review → public-law standing.
AI Verified source (Sep 12, 2026)

added 29d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Richard Blumenthal
U.S. Senator from Connecticut

I am this person! Add a sourced quote Report comment

votes For and says:

In the face of a stunning failure, OpenAI appears to be taking steps that prioritize the performance and profit of its A.I. models with the knowledge that those changes could be detrimental to public safety. This demonstrates the need for vigorous, m...
more AI Verified source (Sep 9, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Yoshua Bengio
AI Pioneer, Turing Award winner

I am this person! Add a sourced quote Report comment

votes For and says:

It is really important, given the stakes, that people who develop AI and those who test it don’t have relationships with each other.
AI Verified source (Sep 1, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Marius Hobbhahn
CEO and co-founder of Apollo Research; AI safety researcher specializing in scheming and pre-deployment evaluations of frontier AI systems

I am this person! Add a sourced quote Report comment

votes For and says:

The current regime is definitely not optimal. While third-party evaluators try hard to be unbiased, a lab can decide to stop working with them any day for any reason whatsoever. We really need more stringent regulations to improve the incentives for ...
more AI Verified source (Sep 1, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Brad Carson
Americans for Responsible Innovation president

I am this person! Add a sourced quote Report comment

votes For and says:

By establishing robust safety standards for developers, requiring independent assurance of those standards, and ensuring transparency into frontier AI development, the federal government can give the public meaningful protections while preserving Ame...
more AI Verified source (Aug 10, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
OpenAI Global Affairs
OpenAI's public policy team.

I am this person! Add a sourced quote Report comment

votes For and says:

At the federal level, our Frontier Safety Blueprint calls for the federal government to require annual independent audits of frontier developers’ safety frameworks and independent third party assessments of the most advanced models, including to moni...
more AI Verified source (Aug 7, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Nathan DeBardeleben

I am this person! Add a sourced quote Report comment

votes For and says:

This paper therefore treats autonomous AI governance as a problem of infrastructure reliability. Its constructive proposal is engineered heterogeneity: the process that proposes an action should not serve as its sole approver and auditor. Independent...
more AI Verified source (Aug 4, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Keivan Navaie

I am this person! Add a sourced quote Report comment

votes For and says:

Building on a structured synthesis grounded in a preregistered bibliometric baseline, we identify five cross-cutting gap dimensions in current alignment research, including the near-absence of institutionalised independent verification. To address th...
more AI Verified source (Jul 27, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
J.B. Branch
AI Governance and Technology Policy Counsel at Public Citizen.

I am this person! Add a sourced quote Report comment

votes For and says:

Congress must establish a comprehensive federal AI safety framework with mandatory safety testing, independent audits, incident reporting requirements, and regulators empowered to remove AI systems when they pose risks.
AI Verified source (Jul 27, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Fathom
Independent nonprofit developing AI-governance and independent-verification frameworks.

I am this person! Add a sourced quote Report comment

votes For and says:

The bill gets the fundamentals correct. It sets one public standard – the adequate mitigation of catastrophic risk – and holds both the AI companies and their independent verifiers accountable to it.
AI Verified source (Jul 23, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Andrew Freedman
Co-Founder and CEO of Fathom, a nonprofit focused on independent AI verification and governance.

I am this person! Add a sourced quote Report comment

votes For and says:

What will work is a competitive market of independent verifiers who are accountable for real-world outcomes and who the government can actually count on.
AI Verified source (Jul 23, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Scott Franklin
U.S. Representative from Florida; member of the House Appropriations Committee

I am this person! Add a sourced quote Report comment

votes For and says:

By tying coverage to substantial AI research and development investments and requiring transparency and independent assessments from those developers, the bill avoids one-size-fits-all mandates and allows smaller innovators to spend their time buildi...
more AI Verified source (Jul 23, 2026) 1 of 2

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Greg Casar
U.S. Representative from Texas; Chair of the Congressional Progressive Caucus

I am this person! Add a sourced quote Report comment

votes For and says:

We need regular mandatory independent safety testing and oversight, mandatory disclosure of security incidents, and international cooperation to keep people safe from absolute disaster.
AI Verified source (Jul 23, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Jay Obernolte
U.S. Representative from California

I am this person! Add a sourced quote Report comment

votes For and says:

Artificial intelligence is advancing rapidly, and Congress must ensure our regulatory framework keeps pace without undermining American innovation. The FRONTIER Act focuses oversight on the largest developers and most advanced models, requiring trans...
more AI Verified source (Jul 23, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Olivier Sibony

I am this person! Add a sourced quote Report comment

votes For and says:

If we place our trust in these products or machines, it is not because they are simple, transparent or harmless. It is because, over decades, we have built a genuine infrastructure of trust around them: laws, regulations, oversight authorities, certi...
more AI Verified source (Jul 22, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
SaferAI
Nonprofit organization researching and promoting responsible frontier AI risk management and safety governance.

I am this person! Add a sourced quote Report comment

votes For and says:

These frameworks are now increasingly being used as a basis for external accountability mechanisms and are turning into regulatory infrastructure.
AI Verified source (Jul 15, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Jack Clark
Anthropic co-founder and Head of Public Benefit

I am this person! Add a sourced quote Report comment

votes For and says:

At this point, everyone at the frontier of AI agrees that third parties should test out AI systems and use these to develop standards to feed into policy.
AI Verified source (Jul 16, 2026) 1 of 3

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Chris Lehane
OpenAI chief global affairs officer

I am this person! Add a sourced quote Report comment

votes For and says:

Companies developing the most capable systems should meet clear requirements, including independent audits, incident reporting, strong security standards, and whistleblower protections.
AI Verified source (Jul 15, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Demis Hassabis
Nobel laureate, AI Researcher and CEO of DeepMind

I am this person! Add a sourced quote Report comment

votes For and says:

Frontier labs would initially share their models with the body voluntarily, up to 30 days before release, for safety testing that probes dangerous cyber, biological and deception capabilities.
AI Verified source (Jul 14, 2026) 1 of 3

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Iqbal Mohamed
UK Member of Parliament for Dewsbury and Batley; former pharmaceutical regulatory professional.

I am this person! Add a sourced quote Report comment

votes For and says:

Secondly, the Government must mandate independent statutory testing for AI before its release, and a yellow card scheme that would be a national register where safety incidents, near misses and harms caused by AI were reported, tracked and acted on e...
more AI Verified source (Jul 8, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Philippines National Police
National police service of the Philippines, represented at the 2026 UN Chiefs of Police Summit.

I am this person! Add a sourced quote Report comment

votes For and says:

Deployment must be accompanied by published guidelines and independent audits.
AI Verified source (Jul 8, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Brendan Steinhauser
CEO of The Alliance for Secure AI; longtime Republican political strategist and former Tea Party organizer

I am this person! Add a sourced quote Report comment

votes For and says:

Governor Pritzker deserves credit for signing this important bill into law, which was passed with strong bipartisan support. Illinois is now the first state to require independent, third-party audits of the largest AI developers’ safety practices. Wi...
more AI Verified source (Jul 8, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Willemien Arends
Minister for Digital Economy and Sovereignty of the Netherlands, speaking on AI governance at the UN Global Dialogue in 2026.

I am this person! Add a sourced quote Report comment

votes For and says:

For example, if we make it more tangible and to have some, some oversight and some control of what's going to happen, we have in the Netherlands the Algorithm Framework that translates legal requirements into practical guidelines also for public orga...
more AI Verified source (Jul 7, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate

Loading more opinions...

Abstain (0)

Loading more opinions...

Against (21)

Paul Lekas
SIIA public policy executive

I am this person! Add a sourced quote Report comment

votes Against and says:

You’ll see industry oppose outside auditors and the main reason is they would mandate something that can’t be done [today] in a robust and uniform manner.
AI Verified source (Jul 8, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate

Loading more opinions...

More h/ai-governance votes

Ban governments from using AI to conduct mass surveillance of citizens

Ban predictive policing

Establish a UN-led body to oversee compute-intensive AI

Mandate disclosure of AI-generated political advertising

Ban fully automated hiring decisions without human review

Back