What Enforcement Patterns in Early AI Stamping Cases Should Change About Firm Onboarding and Training

Licensed engineers can stamp AI-produced engineering drawings only when they exercise the independent professional judgment and responsible charge the seal requires. Approving AI-produced structural calculations does not transfer responsibility to the tool. Enforcement shows AEC firms that onboarding must teach engineers to check and record their engineering decisions as well as use AI software faster.

Why the professional seal has always been an accountability mechanism, not a formality

The professional seal is a legal declaration that a licensed professional exercised judgment and took responsibility for the work. It is not simply a signature on a completed deliverable. AI changes how that deliverable is produced, but it does not change the declaration attached to it. The distinction is between producing engineering work and accepting professional responsibility for its conclusions.

The American Society of Civil Engineers states that AI cannot replace the professional judgment of a licensed Professional Engineer. The NSPE defines responsible charge as requiring engineering decisions to be personally made by the PE “or by others over which the professional engineer provides supervisory direction and control authority”. A tool cannot become the party in responsible charge because it completes a larger share of the production process.

The firm therefore retains responsibility when its licensed engineers sign and seal AI-produced drawings. Neither the apparent completeness of a drawing set nor the speed of its production establishes that responsible charge occurred. The seal represents the professional’s judgment over the engineering decisions underneath those drawings. Because that obligation has not changed, the question for firm leaders is whether onboarding prepares engineers to meet it.

Can engineers trust and stamp AI-produced work without rubber-stamping it?

Rubber-stamping occurs when approval substitutes for substantive professional involvement. In AI-assisted work, the failure is an engineer accepting a generated design without the capacity to independently verify its engineering decisions. Checking a finished deliverable only after the work is done, without the supervisory direction and control that responsible charge requires, amounts to plan stamping.

An engineer may inspect drawings and approve a file without testing whether the reasoning embedded in the design is sound. If the engineer relies on the AI’s conclusion to establish that same conclusion’s validity, the review has no independent foundation.

A policy brief describes a related concern in administrative rulemaking, where AI-generated material can anchor human decisionmakers to faulty reasoning. Its subject is not AEC practice, and its legal context should not be treated as equivalent to engineering licensure. The behavioral parallel is nevertheless direct. Even draft material can inhibit independent reasoning when a decisionmaker accepts its framing without examining the basis for it.

In engineering, that pattern turns an aid into an unexamined authority. Reflexive reliance on AI weakens the review intended to catch the errors it sometimes produces. A PE who cannot independently verify an AI-produced structural calculation cannot make up for that by spending more time looking at the same output. The ethics violation arises when the engineer seals work without exercising the judgment that the seal represents.

What early enforcement and regulatory guidance have actually signaled to the profession

The record described here names no disciplinary case in which a PE stamped AI-produced drawings. The record does show enforcement against misleading AI claims, professional guidance that holds people accountable for their work, and insurance restrictions on AI-related exposure. These signals are relevant without being interchangeable.

The Federal Trade Commission’s Operation AI Comply targets deceptive or unfair conduct involving AI, including misrepresentation of AI capabilities. Its relevance to engineering is narrower than a licensing precedent. AI use does not turn an unsupported representation into an excusable technology error. Firms remain accountable for what they claim a tool can do and for how they rely on those claims.

The FTC has also extended enforcement to B2B marketing and vendors that supply downstream sellers with materials making unsupported AI capability claims. The “means and instrumentalities” doctrine addresses that enabling conduct. AEC firms therefore cannot take training materials from a vendor as neutral evidence that a tool is reliable. Onboarding that repeats exaggerated claims may compound exposure by teaching engineers to rely on capabilities the product cannot support, although that concern is not itself an engineering disciplinary finding.

Insurance creates a separate source of pressure. Verisk CG 40 47 and CG 40 48 introduced generative AI exclusion endorsements on the U.S. commercial general liability form effective January 1, 2026. Individual carriers are also filing AI exclusions for architects’ and engineers’ professional liability coverage at renewal. The two coverage settings should not be conflated. Where continued coverage is conditioned on licensed review, verification, and stamping, the firm’s ability to demonstrate those activities becomes consequential.

That makes training and governance records relevant to underwriting as well as internal administration. A firm that cannot show that independent review took place may face coverage conditions or exclusions when a claim involves AI-assisted work. At the same time, firms may eventually fall short of the standard of care if they turn down AI tools they could use to make work safer and more efficient. Mastery under professional standards is the defensible position, rather than either uncritical adoption or blanket abstention.

The specific gap enforcement has exposed in how firms currently onboard engineers onto AI tools

The central onboarding failure is training directed at the wrong competence. A program can teach engineers to operate an interface, complete a workflow, and move output into project delivery without teaching them to verify the engineering conclusions involved. Operational fluency then becomes a poor substitute for professional readiness.

That gap is especially significant when training remains confined to individual tasks. An engineer may understand how to generate an output without understanding how its assumptions affect the broader engineering decisions on which the project depends. Speed within one workflow does not establish competence across those dependencies.

The ACEC October 2025 AI Adoption Primer describes AI adoption as a change management effort that affects people, processes, and how a firm is organized. When a firm introduces AI as a faster way to perform existing tasks, it can leave engineers’ review responsibilities implicit. Engineers learn what has changed in production without learning what their unchanged professional obligation requires in the new workflow.

Gradon carries out full workflows inside Civil 3D, building information modeling software, and other software firms already use while adapting to each firm’s standards. Those capabilities make the distinction between workflow fluency and engineering judgment especially relevant. Familiar software and firm-specific outputs do not establish that an engineer understands how to verify the decisions embedded in the work. Training centered on speed can encourage confidence without developing the capacity that makes confidence warranted.

The resulting exposure is organizational as well as individual. A firm may have records of who attended training and successful demonstrations of the tool without proof that its engineers can carry out independent review. Its onboarding has prepared people to produce work, but not necessarily to accept responsibility for it. In the current enforcement and insurance environment, that training gap creates an accountability exposure.

What onboarding must actually cover to meet the accountability standard enforcement is making real

A rebuilt program must teach engineers how each tool can fail, how to verify engineering conclusions on their own, and how to document that verification. These distinct capabilities go beyond software proficiency. Together, they connect the use of AI to the responsibility assumed when an engineer approves its work.

Failure-mode training must be specific to the firm’s workflows and deliverable types. Broad warnings that AI can make mistakes do little to help engineers decide where to look more closely. A tool that performs well on clash detection may still anchor an engineer to faulty reasoning on structural load calculations. Competence in one application does not establish reliability in another. Training must explain where the particular tool’s outputs are reliable and where they are not.

Independent verification requires engineers to reach the engineering conclusion, or find its error, without starting from the belief that the generated answer is right. Structured exercises should therefore have engineers compare AI-produced outputs with first-principles calculations or similar manual methods. The objective is the ability to distinguish a sound conclusion from a plausible-looking one and explain the basis for that distinction.

Documentation must make that substantive engagement visible. A timestamp records that approval occurred, but it does not demonstrate the judgment behind the approval. Engineers should be trained to produce a review record that can withstand a licensing board or insurance inquiry. Transparency about AI use must therefore be built into the engineer’s workflow, rather than remaining a general leadership commitment.

An accountability architecture can support that training, as illustrated by Gradon’s model of learning a firm’s past projects and standards before starting work and requiring its licensed engineers to review and approve everything it produces. Working inside existing tools can make AI part of the established production environment. It does not remove the verification obligation. The training still has to establish that required approval means independent engineering review even when an output follows a familiar format.

Barge Design Solutions’ HaSP Assistant is another example, with a focused use, defined inputs, and a review that checks its work against known regulatory requirements. It reduced health and safety plan creation from hours of work to 10–15 minutes. The instructive feature is not speed alone. A reviewer who understands what the tool is doing can test its output against an established requirement, rather than treating the generated plan as its own authority.

Why onboarding alone is insufficient without firm-level governance that enforces the accountability standard over time

Familiarity can erode the caution established during onboarding. As engineers become comfortable with AI tools, reflexive reliance can increase, weakening the independent review they were taught to perform. Completing training therefore does not show that engineers keep meeting the accountability standard on later projects. Governance must sustain the behavior.

The ICE’s updated code extends its treatment of professional conduct to disruptive technologies such as AI and gives guidance on maintaining integrity and accountability when using them. For firms, the practical requirement is a standing workflow in which independent review and documentation remain expected parts of delivery.

The ACEC primer identifies change management and staff ownership as critical to successful integration. Deltek’s Clarity A&E Industry Study identifies insufficient change management and unassigned ownership as barriers that can strand adoption at the pilot stage. Without someone who owns this work, the firm has no lasting way to uphold the review standard as tools and working habits change.

Continued provider involvement can support that structure when it includes working with engineers, tracking tool use, and updating training as the product evolves. Professional responsibility nevertheless remains with the firm and its licensed engineers.

For firm principals, the governance record must show how AI-assisted work is managed across projects and over time. Training records tell part of that story, while review records kept over time show whether engineers put what they learned into practice. The enforcement environment is asking firms to adopt AI in a way that makes professional accountability demonstrable, beginning on the first day an engineer uses the tool and continuing through every approval that follows.