AI Communications, Privilege, and Virginia Workers’ Compensation Practice

Artificial intelligence is no longer a distant technology issue for lawyers. It is already part of how clients, employers, adjusters, claims professionals, medical providers, and injured workers gather information and make decisions. A recent decision discussed in Virginia Lawyers Weekly (“VLW”), United States v. Heppner, addresses an issue that is likely to appear with increasing frequency in civil litigation and administrative claims: whether communications with a public artificial intelligence platform are protected by the attorney-client privilege or work-product doctrine. The court’s answer, on the facts presented, was no.

Although Heppner did not arise in the workers’ compensation context, the issue has practical importance for Virginia workers’ compensation practice. Claims are often investigated and evaluated through written communications among injured workers, supervisors, human resources personnel, adjusters, nurse case managers, medical providers, and counsel. As generative AI becomes more common, some of those individuals may use public AI tools to summarize facts, evaluate compensability, draft claim notes, prepare timelines, analyze medical issues, or organize hearing strategy. Those exchanges may not be privileged simply because they relate to a legal dispute.

The Decision

In Heppner, the defendant argued that his communications with Claude, a public AI platform, were protected because they concerned anticipated legal defenses and litigation strategy. The court rejected that argument. The communications were not between attorney and client, the AI platform was not acting as counsel, and the defendant had not shown that the exchanges were confidential in the manner required for privilege protection. The court also declined to apply work-product protection where the communications were not prepared by or for counsel.

The decision did not hold that all AI-assisted legal work is unprotected. A different analysis may apply where counsel directs the use of an appropriate tool under conditions that preserve confidentiality and where the work is performed for purposes of legal advice or anticipated litigation. The decision instead illustrates the risk of unsupervised use of public AI tools in connection with a pending or anticipated legal matter.

There is also, at least for now, some disagreement among courts on this issue. As noted in the VLW article, a Michigan court reached a different conclusion and found certain AI-related communications privileged. That distinction should not be ignored. However, the safer view for Virginia practitioners is that the New York decision reflects the more conservative approach courts are likely to take, particularly where the AI tool is public-facing, the communications were not made at counsel’s direction, and the platform’s terms do not ensure confidentiality. Until there is clearer authority, Virginia workers’ compensation practitioners should assume that public AI communications may be discoverable unless the use is structured and supervised in a manner that satisfies traditional privilege and work-product requirements.

Why the Issue Matters in Workers’ Compensation Claims

Virginia workers’ compensation claims often develop before counsel is involved. An employee reports an injury. A supervisor prepares an accident report. Human resources gathers information. An adjuster evaluates compensability. A claimant searches for benefits information. A nurse case manager or claims professional summarizes medical records. At any point, one of those individuals may now use a public AI tool.

That use may create a discoverable record. A claimant may ask an AI platform whether a preexisting shoulder condition can still be claimed as a work injury. An employer representative may ask whether a claim can be denied despite an accident report that supports the employee’s version of events. An adjuster may ask for a denial rationale based on a specific set of facts. These prompts may contain factual admissions, inconsistent statements, or claim evaluations that would not otherwise have been written down.

The output from the AI platform may also matter. An AI-generated chronology, compensability assessment, medical summary, or settlement analysis may reflect the user’s factual assumptions and theory of the case. If the material was created outside counsel’s direction and without adequate confidentiality protections, a party may have difficulty asserting privilege or work-product protection.

Potential Discovery Implications

Discovery in Virginia workers’ compensation claims is more limited than traditional civil discovery, but the Commission permits discovery relevant to disputed issues. As AI use becomes more common, parties may begin asking whether AI tools were used to prepare factual summaries, accident timelines, claim evaluations, medical chronologies, job search materials, hearing statements, or settlement analyses.

The issue is most likely to arise where there is already a factual dispute. Examples include contested accident mechanics, preexisting conditions, delayed reporting, inconsistent medical histories, refusal of selective employment, willful misconduct, intoxication, horseplay, termination for cause, or disputed marketing efforts. In those cases, AI prompts and responses may contain statements bearing on credibility, causation, disability, or defenses.

Requests for AI-related materials should still be targeted. The mere fact that a person used AI does not make every exchange relevant or discoverable. The better inquiry is whether the AI communication contains claim-specific facts, admissions, inconsistent statements, medical assumptions, defense analysis, or materials used to prepare testimony or claim submissions.

Employer, Carrier, and TPA Use

From the defense perspective, the most immediate risk is client-side use. Employers, carriers, and third-party administrators may assume that entering information into an AI platform is no different from preparing an internal note. That assumption may create problems.

Claim-specific facts, medical information, personnel information, witness statements, settlement authority, and defense assessments should not be entered into public AI tools. This is especially important before counsel is involved, when the user may not be thinking in terms of privilege, work-product protection, or discovery. It is also important in claims involving sensitive medical records, personnel issues, or disputed factual allegations.

Counsel should consider giving this instruction early in the representation. Employers and claims professionals should be told not to use public AI tools to evaluate compensability, draft denial rationales, summarize medical records, or prepare litigation strategy using claim-specific information unless counsel has approved the tool and the process.

Claimant Use

The same issue may arise on the claimant side. Injured workers may use AI to research benefits, draft statements, summarize medical records, explain gaps in treatment, or prepare for hearing. Some of that use may be ordinary research or writing assistance. Some of it may become relevant if the prompts or outputs contain inconsistent facts, admissions about non-work causes, descriptions of symptoms, or explanations prepared for use in the claim.

For example, if a claimant asks an AI platform how to connect a degenerative condition to a minor work incident, the communication may be relevant to causation and credibility. If a claimant uses AI to draft a job search log or explain a refusal of light duty, the underlying AI exchange may be relevant depending on how the material was used. The analysis remains fact-specific.

Counsel’s Use of AI

Current guidance indicates that lawyers may use AI responsibly, but the privilege analysis depends heavily on the tool and the workflow. Public consumer platforms present different risks than legal-specific or enterprise tools with contractual confidentiality protections, restricted data use, and appropriate retention limits. Before entering claim-specific information into any AI tool, counsel should understand who can access the data, whether inputs are retained, whether the information may be used to train the model, and whether the provider can disclose the information to third parties.

Counsel should also verify all AI-generated work. This is particularly important in workers’ compensation practice, where small distinctions in medical causation, disability, notice, marketing, selective employment, and procedural deadlines can determine the outcome. AI may be useful for organizing records or preparing a first draft, but it cannot substitute for legal judgment, medical-record review, or verification of Virginia statutes, Commission rules, and controlling authority.

Practical Guidance

In order to anticipate some of the issues discussed above, Virginia workers’ compensation practitioners may consider implementing the following practical steps.

First, counsel should instruct employers, carriers, TPAs, and other claim stakeholders not to place claim-specific information into public AI tools. That instruction should include medical records, accident facts, witness statements, personnel information, settlement authority, and defense assessments.

Second, counsel should ask early whether AI has already been used in connection with the claim. If it has, counsel should determine what tool was used, who used it, what information was entered, what output was generated, and whether any output was saved, shared, or relied upon.

Third, discovery should be updated where appropriate. In cases involving disputed facts or credibility issues, targeted discovery regarding AI-generated factual summaries, timelines, statements, medical chronologies, or claim evaluations may be appropriate.

Fourth, any AI use by counsel should be structured to preserve confidentiality and work-product protection. The use should be directed by counsel, performed through an appropriate platform, and limited to tasks that counsel supervises and verifies.

Finally, lawyers should treat AI-generated material as a draft or organizational tool, not as an authority. Any legal citation, statutory reference, medical summary, or case description generated by AI should be independently checked before it is relied upon.

Conclusion

Heppner does not create new rules for AI when it comes to privilege. It applies traditional privilege and work-product principles to a new form of communication. Attorney-client privilege still requires confidential communication made for the purpose of legal advice. Work-product protection still depends on materials prepared by or for counsel in anticipation of litigation. Public AI communications may simply fail those requirements.

For Virginia workers’ compensation practice, the lesson is practical. AI use should be addressed at the beginning of a claim, not after a discovery dispute arises. Used carefully, AI may help organize information and improve efficiency. Used casually, it may create discoverable statements, weaken privilege arguments, or expose sensitive claim analysis.

Posted In: E-Blast