A research team at Stanford Law School ran a benchmark study on general-purpose AI tools last year and found hallucination rates between 58 and 88 percent. That number landed quietly in the legal press. It is landing very loudly inside the professional liability market for architects and engineers.

The standard of care for a design professional is not new, and it has not changed. A reasonably prudent architect or engineer, performing services under similar circumstances, exercising the skill and judgment ordinarily expected of the profession. Courts have been applying that definition for decades. Licensing boards apply it. AIA Document B101 codifies it. ASCE policy reinforces it. The professional who stamps or seals the work is responsible for verifying accuracy — regardless of how the work was produced.

The piece that is new is the tool. And the tool changes the operational picture in ways the standard does not bend to accommodate.

Three things moving at the same time

First, the carriers stopped waiting for a court ruling. Verisk's AI exclusion endorsement went into effect January 1, 2026. AIG, Great American, and WR Berkley have filed for regulatory approval to add AI exclusions to E&O forms. Philadelphia Insurance and Hamilton Select have already excluded AI-related claims. That is not a future risk. That is a current endorsements page on most renewal proposals being written this quarter. The COI never changes. The endorsements page changes every year. A firm that reads only the COI is reading the wrong document.

Second, the AI exclusions are not uniform. Some carriers are using broad-form language that excludes any claim “arising from the use of artificial intelligence.” Others are narrowing to “autonomous AI decision-making.” Others are carving out narrow defense-cost coverage with no indemnity. A firm that uses ChatGPT to draft a specification, a Copilot to outline a report, or a diffusion-based tool to render a concept is not doing the same thing as a firm that runs a generative AI tool through a structural analysis loop. The policy form needs to map to the actual workflow. The COI does not show that.

Third, the legal standard the plaintiff will argue is unchanged. The plaintiff does not have to prove the AI was negligent. The plaintiff has to prove the engineer of record did not meet the standard of care. If the standard of care includes verifying that an AI-generated output is accurate — and every published professional society guidance now says it does — then the duty to verify is the duty the policy is being asked to defend. The exclusion lands on the wrong side of that duty.

This is the gap. And it is widening fast in 2026.

Coverage is now a function of form, not carrier name

There is a quieter pattern underneath all three. The traditional broker model in A&E has been certificate-driven. Renew the E&O. Confirm the limits. Send the COI to the client. Move on. That model produced a culture where firms believed coverage was a function of carrier name and policy limit. The 2026 market is making it brutally clear that coverage is a function of form, of endorsement, of workflow disclosure, and of operational discipline around the actual tools the firm uses every day.

PFTN was built to be the opposite of the certificate broker. Our 4-Step Strategic Process for design professionals starts with Strategic Discovery: project types, jurisdictions, AI use cases by workflow, BIM stack, jurisdiction-specific stamp-and-seal requirements, prime-contract indemnity exposure. Risk Assessment quantifies what most brokers never look for: actual policy form, AI exclusion language, cyber-event seam, attachment points on the umbrella, and the gap between the firm's written AI use policy and the renewal questionnaire the carrier sends. Solution Design pairs the right E&O form with the right cyber form, and — when it matters — a contractual liability or technology E&O placement that closes the AI workflow seam the standard forms create. Ongoing Optimization re-checks the form every year, not just the limits.

There is a deeper version of this same point.

AI does not change the standard of care. It changes the work the design professional has to do to meet it. The verification step is no longer optional. It is the entire job. The firm that builds a written AI use policy, documents the verification workflow, trains the project teams, and sits down with the broker to map the policy form to the actual workflow — that firm is in a fundamentally different conversation than the firm that renews on autopilot.

The plaintiff's bar is reading the same Stanford report the underwriters are reading. The discipline that wins this year is the discipline that engineers the protection, not just the design.

The shift starts with one conversation — and preferably long before the next renewal hits.

— Ryan Mefford, President & Risk Advisor