AI Is Now Central to Claims Work. Here's How to Use It Without Creating New Risks
Generative AI has moved from the margins into everyday claims workflows. Adjusters use AI-assisted tools to sort documents, summarize files and support litigation strategy. Defense counsel uses the technology to draft motions, organize chronologies and prepare initial case assessments.
The efficiency gains are real: faster document review, shorter cycle times and more consistent case valuations. But for risk managers overseeing claims operations, the central question has shifted. It's no longer whether to use AI, but how to deploy it without introducing operational, legal or reputational risks.
Where AI delivers tangible value
AI proves most useful in three core areas of claims handling: document review, case summarization and litigation support.
Document review. In complex casualty or coverage disputes, claims teams face hundreds or thousands of pages of discovery, medical records, correspondence and regulatory filings. AI tools can quickly flag key documents, identify patterns like inconsistent medical histories or contradictory statements, and prioritize materials that may influence liability or damages.
This frees adjusters to focus on higher-level work: negotiating settlements, working with injured parties and managing mediation. AI also handles the practical challenge of large volumes of scanned files, turning static PDFs into searchable information that makes it easier to build timelines and connect facts.
Case summarization. AI generates concise summaries from pleadings, deposition transcripts, expert reports and internal emails. Underwriters and risk managers use these summaries to quickly assess exposure, identify coverage issues and monitor portfolios at scale.
For multistate portfolios, this capability reveals jurisdiction-specific patterns: rising punitive-damages demands in certain states, or shifting judicial attitudes toward policy interpretation. Risk managers can then adjust underwriting, reinsurance or claims-handling strategies proactively.
Litigation support. Defense counsel uses generative AI as a drafting tool. Attorneys and paralegals use it to outline motions, draft discovery responses and prepare settlement demand letters. Some firms use AI to generate preliminary briefs or organize deposition exhibits and timelines.
Some insurers have integrated AI-driven support into panel management workflows, comparing briefs across similar cases and estimating typical settlement ranges based on past results.
The risks: Hallucinations, bias and over-reliance
Three major concerns dominate AI risk management in claims work: hallucinated legal citations, built-in biases and over-reliance on AI-generated analysis.
Hallucination. AI systems produce false citations, inaccurate summaries and confident-sounding explanations that are simply wrong. A 2023 federal case from the Southern District of New York involved lawyers who relied on ChatGPT-generated case citations that did not exist, resulting in sanctions.
In March 2024, the U.S. District Court for the Middle District of Florida disciplined an attorney for filings that included fabricated case law. These cases reinforce that responsibility for accuracy remains with counsel, not the tool.
Bias. Models trained on historical claims data may reflect past valuation patterns, jurisdictional tendencies or demographic disparities. In workers' compensation or disability claims, this could unintentionally skew estimated payouts or settlement outcomes.
Over-reliance. If adjusters treat AI-generated exposure assessments or liability predictions as definitive, they may overlook data gaps or flawed assumptions. This can result in incomplete or error-laden outcomes, regulatory scrutiny or reputational harm.
Professional guidance and ethics
Ethics bodies and courts are clarifying expectations. The American Bar Association's Formal Opinion 512 treats generative AI as a form of nonlawyer assistance, requiring lawyers to maintain competence, supervise outputs, protect confidentiality, communicate with clients and ensure fees remain reasonable.
State-level guidance is developing along similar lines. Florida Ethics Opinion 24-1 permits AI tools with appropriate safeguards. Delaware adopted an interim policy in 2024 for judicial officers and court personnel, restricting use of nonpublic data in unapproved AI tools and emphasizing that decision-making must remain with humans.
For claims organizations, these principles translate into clear governance requirements: vet and approve designated tools, require human validation of outputs and maintain documented oversight processes.
Data security and confidentiality
Claims files contain medical records, wage information, employment history and other sensitive data. Many AI tools require users to upload material to cloud-based systems, raising questions about retention, training and unauthorized sharing.
Risk managers should not assume every vendor handles data the same way. Contracts should address encryption, retention, access controls and whether submitted materials may be used to train the model. Where possible, require tools that do not store data or use anonymized data for higher-risk situations.
This matters especially in litigated claims, where sensitive records may later be scrutinized in discovery or by regulators. Weak data governance can turn a productivity tool into a liability.
Emerging regulatory attention
Legislative and regulatory attention to AI risk is increasing. Proposed federal legislation such as the AI LEAD Act (S.2937) would subject AI developers to liability like that faced by product manufacturers, including for defective design or failure to warn.
For insurers, third-party administrators and defense firms, this evolving landscape raises important considerations. Organizations that customize or rely heavily on AI tools may face scrutiny over how those tools are used and whether adequate oversight is in place.
Practical risk controls
Treat AI like any other high-impact operational system: with written rules, testing and accountability.
- Adopt clear use policies. Define what AI may and may not do, require human review of all outputs used in legal or claims decisions, and set training expectations for users.
- Verify AI-generated work independently. Any citation, fact statement or settlement recommendation should be checked before relying on it. This safeguard is especially important given recent disciplinary actions.
- Review vendors as part of governance. Claims leaders, counsel and technology teams should understand how a tool is trained, what data it uses, how bias is tested and how updates are handled.
- Monitor performance over time. Periodic audits comparing AI results against human review can reveal whether the system is drifting, reinforcing bias or missing key issues. Document errors and correct them promptly.
Moving forward
Generative AI is here to stay in claims management. Recent cases, ethics guidance and emerging legislation all point toward a future where AI-assisted work is treated with the same level of scrutiny as any other form of professional judgment.
The most resilient organizations will insist on human oversight at every critical decision point, embed AI governance into broader risk-management and compliance frameworks and treat AI vendors as partners in risk management, not just cost-saving tools.
For risk managers, the question is no longer whether to use AI. It is whether they can implement it in a way that balances efficiency with accountability and innovation with disciplined risk management. Learn more about AI for Insurance and AI for Legal Professionals to build competency across your organization.
Your membership also unlocks:
