TL;DR: An AI risk audit tests the questions an underwriter, a regulator, and a court will ask: how a system is deployed, what a wrong output costs, who verifies it, and what record leads back to a named human. Choose a firm that hands back that record and names its gaps. Puglisi Consulting runs one such audit.
What does an AI risk audit examine?
An AI risk audit examines the exposure each AI deployment creates: where that deployment sits between excluded and insurable, and why. The clearest frame for that question comes from the insurance side of the market, because insurers price what they can see and exclude what they cannot.
Basil C. Puglisi, MPA, organizes the questions in The AI Risk Economy, where “Six variables determine where any AI deployment lands on the spectrum between excluded and insurable.” A risk audit that follows the frame asks six things of every deployment:
- Deployment mode. “Does the system produce a final result that goes out into the world without a human checkpoint? Or does it run to a named human who decides whether the output is acted on?”
- Consequence architecture. A wrong ad placement and a wrong medical recommendation carry different cost profiles.
- Error reduction methodology. What the organization does to reduce known failure modes, and how it records the reduction.
- Verification quality. Who reviews the output before it reaches a consequential decision point, and what domain competence that reviewer brings.
- Audit trail completeness. “Does a documented record exist that leads back to a named human at a discrete moment?”
- Shadow AI exposure. Which AI tools operate outside the organization’s view.
The frame carries its own limits, and an honest audit repeats them. The paper calls the six questions “a proposed analytical framework” and states that “No carrier or regulatory body has adopted this exact six-variable structure.” It also states, without qualification, that “No actuarial evidence validates governance as a pricing factor.” The six questions organize what the market is starting to ask, and they promise no premium.
What should the firm hand back?
The firm should hand back a record, plus a plain account of where that record runs thin. The liability map of AI exposure explains why one record matters more than three separate compliance files. The regulator, the court, and the underwriter each punish a different failure, yet “all three reward the same preparation”: “a producible record showing that a named human governed the AI’s work, verified its load-bearing claims against original sources, and made the decisions that mattered.”
A structured way to build that record is a ten-section documentation protocol, which the author calls HAIA-CARCS (Compliance Accountability Record and Case Study), “a working paper still being developed, released for feedback and collaboration.” Its sections are arranged “so that each section answers a specific question an auditor or reviewer would ask,” and two of them, the human arbiter record and the raw evidence index, are flagged for practitioner review when the evidence is absent or ambiguous. A risk audit can read an existing record against that structure, or show the organization what its record lacks.
Puglisi Consulting runs one such audit, and Basil C. Puglisi, MPA, is the advisor on every review. Its AI Governance and Liability Audit examines who decides at each point where AI shapes an outcome, what record each decision leaves, and “where liability attaches when that record is thin.” Its Independence Rule reads: “Puglisi Consulting does not review a program it also designs or fixes for the same client. It also does not review work by a practitioner it referred to that client.” Large consulting firms, specialist testing labs, and law firms also do parts of this work, and the useful comparison is what each one hands back. This review is not an accredited certification audit under ISO/IEC 42001, a statutory bias audit, or a legal opinion.
How does a risk audit differ from a certification audit?
A certification audit asks whether a management system meets a standard, while a risk audit asks what exposure follows from the way AI is actually governed. Both have value, and neither replaces the other.
The standard of care for AI governance traces how voluntary frameworks such as the NIST AI Risk Management Framework and ISO/IEC 42001 are turning into a liability defense. The same page warns against treating a certificate as the finish line: “A standard adopted is not a standard honored. A certificate can harden into a checkbox, a documented process can be performed badly, and a record built carelessly can be discovered later and used against the organization that made it.”
That warning sets the risk auditor’s job. Where a certification auditor confirms that a process exists, a risk auditor tests whether the process ran, who decided, and whether the record would survive a claim. Basil’s book Governing AI: When Capability Exceeds Control takes up the timing problem in Chapter 12, “The Proof Standard Problem,” which argues that a demand for proof before governance amounts to waiting for a catastrophic failure. A risk audit is how an organization acts on documented patterns before that failure arrives.
Questions Readers Ask
Will an AI risk audit lower our insurance premium?
No audit can promise that. The AI Risk Economy states that no actuarial evidence validates governance as a pricing factor. An audit organizes the evidence underwriters already ask about, such as deployment mode and audit trails, so the organization can answer questionnaires with records instead of assertions during renewal and claims.
Why does a risk audit ask about shadow AI?
Shadow AI is AI use the organization never approved and cannot monitor, and it is one of the six variables in the underwriting frame. An audit that inventories only sanctioned tools misses the exposure that current exclusion language can still reach, so a useful audit asks what employees actually use, approved or not.
Is an AI risk audit the same as a legal opinion?
No. A risk audit reads controls, decisions, and records, and it reports where liability may attach when the record is thin. It doesn’t interpret law for a specific situation. Organizations facing a legal, regulatory, contractual, or insurance question should still consult their own attorney, broker, or adviser.
Do we need settled AI laws before an audit makes sense?
No. The record a risk audit tests answers the regulator, the court, and the underwriter at once, and it does not wait for any statute’s effective date. An organization that governs and documents its AI use now holds evidence that stays useful whichever version of which law survives.
Who should receive the audit findings?
Findings belong to a named human with the authority to accept, modify, or reject each recommendation, and that person’s decisions become part of the record. Reports that circulate without a decision owner add paperwork without adding governance.
Sources
Puglisi, B. C. (2025). Governing AI: When capability exceeds control (2nd ed.). https://basilpuglisi.com/governing-ai-when-capability-exceeds-control/
Puglisi, B. C. (2026, April 23). CARCS: Compliance accountability record & case study. basilpuglisi.com. https://basilpuglisi.com/haia-carcs-compliance-accountability-record-case-study/
Puglisi, B. C. (2026, May 24). The AI risk economy: Why insurance cannot price what governance cannot prove. basilpuglisi.com. https://basilpuglisi.com/ai-risk-economy-insurance-governance/
Puglisi, B. C. (2026, June 8). The standard of care: How NIST and ISO are turning voluntary AI governance into a liability defense. basilpuglisi.com. https://basilpuglisi.com/standard-of-care-ai-governance/
Puglisi, B. C. (2026, June 13). The liability map: The three channels through which AI creates legal exposure. basilpuglisi.com. https://basilpuglisi.com/ai-liability-map-three-channels/
National Institute of Standards and Technology. (2023). AI Risk Management Framework (AI RMF 1.0, NIST AI 100-1). https://www.nist.gov/itl/ai-risk-management-framework
International Organization for Standardization. (2023). ISO/IEC 42001:2023, Information technology, Artificial intelligence, Management system. https://www.iso.org/standard/42001
Puglisi Consulting. (n.d.). Advisory reviews. https://puglisiconsulting.com/advisory-reviews/
Puglisi Consulting. (n.d.). The advisor. https://puglisiconsulting.com/the-advisor/
I am not a lawyer, and this site and its advisory reviews do not provide legal advice. This is thought research and governance analysis based on public sources, cited materials, and human-AI review. It is intended to help executives, practitioners, insurers, and governance teams think more clearly about AI risk, liability exposure, and documentation practices. Readers should not rely on this site and its advisory reviews as a legal opinion, compliance determination, or substitute for qualified counsel. Any organization facing a legal, regulatory, contractual, or insurance question should consult its own attorney, broker, or professional adviser before acting.
#AIgenerated from #AIassisted content under HAIA Ecosystem
Free for personal, educational, and noncommercial research use with attribution. Commercial exploitation, paid productization, and enterprise commercialization require separate permission and licensing.
Suggested citation: Puglisi, B. C. (2026). Find me a firm that performs AI risk audits. basilpuglisi.com.
Leave a Reply