Table of Contents

1. Key Highlights:
2. Introduction
3. From Retail Tech to Health Benefits: Why Lavu’s Move Matters
4. Three Priority Pillars: Member Chat, Prior Authorization, and Corporate Productivity
5. How Agentic AI Fits—and Why Guardrails Matter
6. Data, Privacy, and Regulatory Considerations
7. Upskilling and the Human Element
8. Performance, Metrics, and the Business Case
9. Broader Industry Signals: Vendor Risk, Market Shifts, and High-Profile Incidents
10. Clinicians vs. Executives: The Trust Gap
11. Operationalizing Responsible AI: Governance Frameworks and Monitoring
12. Balancing Speed and Safety: When to Automate and When to Wait
13. Real-World Implications: Members, Providers, and Payers
14. External Market and Policy Forces That Will Shape Adoption
15. Lessons from Incidents and Market Tensions
16. Practical Roadmap for Health Organizations Considering AI
17. Where Elevance’s Strategy Could Lead the Industry
18. The Limits of Technology: Why Human Judgment Remains Central
19. Final Considerations: Pace, Proof, and Public Trust
20. FAQ

Key Highlights:

• Elevance Health, under new CDIO Ratnakar Lavu, is deploying generative AI across member-facing apps, prior-authorization workflows, and internal productivity tools—targeting improved access, lower costs, and faster decision-making for 45.2 million medical members.
• The company balances automation with human oversight: pilot deployments (Sydney chat, prior-authorization automation, Spark) show scale—60,000 associates using Spark and 10 million messages in H1 2025—while guardrails, data curation, and certification programs aim to manage safety, privacy, and regulatory risk.
• Broader industry signals—executive enthusiasm, clinician caution, regulatory scrutiny of AI vendors, and high-profile vendor incidents—underscore that health-system transformation depends on responsible models, rigorous testing, and workforce upskilling.

Introduction

Healthcare systems generate immense volumes of data and complex workflows that scramble patient access and inflate costs. Elevance Health, the Fortune 500 health benefits giant that serves tens of millions of people, has chosen to place AI at the center of its strategy to untangle those processes. The move is notable because the company's new chief digital and information officer, Ratnakar Lavu, built his reputation transforming consumer brands at Macy’s, Kohl’s and Nike. His arrival signals a belief that customer-oriented design and data-driven automation can be applied to healthcare at scale—but only if done with carefully defined limits.

Elevance’s experiments span member-facing conversational agents, automated prior-authorization, and enterprise-level employee assistants. Each use case offers measurable efficiency gains and improved member experience potential. They also raise familiar concerns: model accuracy in clinical contexts, data privacy under HIPAA, biased outcomes when training data is imperfect, and the operational risk of agentic systems that perform multi-step tasks autonomously. Lavu’s approach stresses targeted deployments, human supervision, and broad internal upskilling. That combination reflects the pragmatic stance that many healthcare executives are taking: adopt AI where the returns are clear, manage risk through guardrails and workforce training, and watch for unintended consequences.

The rest of this article examines Elevance’s three-pronged AI strategy, the technical and governance challenges it faces, how those plans map to broader healthcare AI adoption trends, and the market forces reshaping enterprise AI partnerships and playbooks.

From Retail Tech to Health Benefits: Why Lavu’s Move Matters

Ratnakar Lavu’s career arc—from retail technology leadership to the C-suite at Nike and Kohl’s, and now CDIO at Elevance Health—looks like an abrupt sector leap on the surface. The connective tissue is user experience and scale. Retail and healthcare both involve large, heterogeneous customer bases and transactions that hinge on clarity, timeliness, and trust. Lavu’s prior roles focused on personalization and seamless consumer journeys; Elevance offers a more constrained but higher-stakes environment in which those same design principles can reduce friction and cost.

Elevance Health reported $197.6 billion in operating revenue for 2025 and serves 45.2 million medical members. A company of that scale can realize significant financial and clinical impacts when automation reduces unnecessary out-of-network referrals, speeds administrative approvals, or prevents errors that generate costly claims. Lavu’s retail instincts—prioritizing simplified flows, contextual recommendations, and personalization—translate directly into member-facing experiences that answer a basic question consumers repeatedly ask: Am I covered, and how much will it cost?

Applying retail-style personalization to health benefits requires translating commercial metrics into clinical and regulatory terms. A suggestion for a shoe is low risk compared to recommending a surgical pathway. Elevance has positioned AI as a facilitator of navigation—helping members find in-network providers, estimate out-of-pocket costs, and access benefits information—rather than as a replacement for clinical judgment.

Three Priority Pillars: Member Chat, Prior Authorization, and Corporate Productivity

Elevance’s AI investments focus on three pillars that span customer experience, insurance operations, and employee productivity. Together they illustrate different risk profiles and technical patterns for AI adoption.

1. Member-facing conversation (Sydney mobile app)

• What it does: A ChatGPT-like conversational layer in Elevance’s Sydney mobile app enables members to ask natural-language questions about coverage and costs—e.g., “I need knee surgery. Am I covered and how much will my co-pay be?” The system maps coverage rules, deductibles, and regional provider networks to deliver actionable answers.
• Why it matters: Members often face confusing plan details and provider networks. A chat interface that translates plan language into plain English and surfaces in-network options reduces surprise bills and makes care choices more cost-effective.
• Risk profile: Moderate. The system must accurately interpret coverage terms and network status, and it must clearly communicate uncertainty and explain when a human review is required.

2. Prior-authorization automation

• What it does: Prior authorization requires clinicians to obtain plan approval before certain treatments or prescriptions. Elevance has automated portions of those workflows using AI to evaluate documentation against coverage policies and clinical criteria. Humans still review borderline or high-stakes cases.
• Why it matters: Manual prior authorization adds delays to care, drives administrative costs for providers, and creates patient frustration. Even partial automation can reduce turnaround times and administrative burden, streamlining access to necessary interventions.
• Risk profile: Higher. Decisions affect whether a patient receives care or medication. Accuracy, clinical-safety validation, and explicit appeal paths are essential.

3. Internal productivity assistant (Spark)

• What it does: Spark is an internally built chatbot that aggregates multiple large language models to support document analysis and common tasks across corporate functions. More than 60,000 Elevance associates used Spark to exchange 10 million messages in the first half of 2025.
• Why it matters: Large organizations contend with knowledge silos, repetitive requests, and slow document reviews. An enterprise assistant that integrates internal knowledge bases, policies, and secure data stores accelerates employee productivity and decision-making.
• Risk profile: Low-to-moderate. Internal tools pose lower clinical risk but must uphold data security, access controls, and guardrails to prevent leakage of member data or inappropriate model outputs.

Each pillar requires a different technical design, compliance posture, and operational oversight. The member chat needs explainability and clear boundaries (informational rather than clinical diagnosis). Prior authorization automation demands clinical validation, auditable decision trails, and an appeals process. Spark must be secured, monitored, and incorporated into employee training so outputs are used responsibly.

How Agentic AI Fits—and Why Guardrails Matter

Agentic AI, systems that can perform multi-step tasks autonomously by creating sequences of actions, holds particular appeal for complex healthcare workflows. Examples include an agent that can research a member’s benefits, contact a provider for missing documentation, submit a prior-authorization request, and monitor the response. Agentic systems promise to compress multi-actor processes into faster, more reliable flows.

Elevance is exploring agentic use cases across its three pillars and expects to scale them more aggressively in 2026. The company’s approach to agentic AI emphasizes two principles:

• Precise task definition: Agents must be trained and constrained only on the specific data needed for their assigned tasks. That reduces the risk that an agent performs beyond its remit or leverages data inappropriately.
• Human oversight: A human can supervise multiple agents, but oversight is mandatory. That oversight includes step-by-step approvals for high-risk decisions, structured exception handling, and periodic audits of agent behavior.

Agentic AI compounds certain risks. When an agent chains together API calls, database queries, and content generation, small errors can cascade. If a model misreads a care plan or misroutes a prior-authorization request, a patient might experience delayed care. Agents must therefore maintain auditable logs, role-based access controls, and explicit stop conditions. Elevance’s emphasis on training agents on narrow datasets and ensuring employee oversight reflects a governance-minded approach that recognizes autonomy increases both value and risk.

Operationalizing oversight requires tooling: dashboards that surface agent decisions, anomaly detectors for agent drift, and automated rollback or quarantine mechanisms when agents behave unexpectedly. Those technical investments are as important as model selection.

Data, Privacy, and Regulatory Considerations

Healthcare AI sits at the intersection of business transformation and strict regulation. HIPAA and state privacy laws constrain how member data can be used, stored, and shared. The risk vector widens when AI vendors host models in cloud environments that may require additional contractual assurances.

Elevance’s use of internal models (Spark) and enterprise ChatGPT licenses indicates a hybrid strategy: build proprietary systems where control and compliance are highest priority, and selectively adopt external platforms under enterprise agreements that provide contractual security and governance features. The company’s decision to offer tens of thousands of employees opportunities to earn AI fluency certifications is a governance move: educated users are less likely to misuse models or expose protected information.

Key data and compliance principles to operationalize:

• Minimum necessary data: Agents and models should be trained and exposed only to the data essential for the task.
• Data lineage and provenance: Every inference should be traceable to input data and model version at the time of decision.
• Encryption and access control: Data in transit and at rest must be encrypted, and access must be role-based and auditable.
• Vendor risk management: Contract terms should address model governance, training data retention, third-party risk, and incident response obligations.

These principles also play into public trust. Gallup polling shows U.S. satisfaction with healthcare costs has fallen to a record low—only 16% satisfied—heightening the stakes for any organization promising efficiency improvements. Transparent governance and clear consumer-facing explanations are necessary to maintain confidence.

Upskilling and the Human Element

Elevance is investing in workforce readiness. The company is rolling out enterprise ChatGPT licences and certification programs in AI fluency that include responsible AI, prompt engineering, and advanced AI-enabled workflows. The goal is to create a baseline of competence so that tens of thousands of associates make better decisions about when to trust AI outputs and when to escalate.

Upskilling accomplishes several objectives:

• Safer deployment: Trained employees know how to validate model outputs, spot hallucinations, and follow escalation protocols.
• Productivity gains: Employees who use AI effectively can process more cases, reducing backlogs and improving responsiveness.
• Cultural acceptance: Hands-on training reduces fear and builds champions who can drive adoption responsibly.

Health systems and payers face a macro workforce shortage: analyses project at least a 10 million global shortfall in healthcare workers by 2030. Executives see AI as a partial remedy—three out of four healthcare executives in a recent survey believed AI would have a positive impact on care quality. Clinicians, however, were more skeptical; only 45% believed AI could address staffing shortages to some extent, and 6% thought it could solve the problem mostly or completely. That gap highlights the need to align technology investments with clinician workflows and to demonstrate improvements in low-stakes settings before expanding into clinical decision-making.

Upskilling programs should be role-specific: clinician-focused modules on model interpretability and safety, claims staff modules on automated adjudication, and compliance modules on data privacy. Certifications should be paired with practical, monitored deployments.

Performance, Metrics, and the Business Case

For AI to become more than pilot projects, health systems must measure and realize tangible returns. Elevance’s use cases map cleanly to measurable KPIs:

• Member chat (Sydney): metrics include resolution rate (questions answered without human escalation), reduction in call-center volume, member satisfaction scores (NPS), and downstream metrics such as reduced out-of-network utilization.
• Prior-authorization automation: primary metrics are average decision time, percent of requests auto-approved, appeals rate, and clinical correctness compared to human adjudicators.
• Spark and workplace assistants: metrics include time-to-respond for internal requests, document-processing throughput, and error rates in document interpretation.

Quantifying cost savings requires modeling labor-time reclaimed, reduction in avoidable claims, and the value of improved member retention. For example, if automation reduces average prior-authorization adjudication from several days to hours, the downstream effect on patient satisfaction and provider willingness to participate in-network can be significant. Similarly, preventing out-of-network referrals through accurate in-network provider recommendations reduces both member out-of-pocket costs and payer expenditures.

Real-world examples illustrate these dynamics:

• Retailer Walmart credits AI for improved inventory decision-making, reduced friction, and automation that shifts freight distribution toward efficiency. The company reports that 60% of stores receive freight from automated distribution centers, demonstrating how automation yields measurable operational gains.
• In healthcare, automation of administrative tasks—documentation and EHR workflows—already shows adoption: 71% of surveyed healthcare entities report AI use for documentation. When mundane, low-risk tasks are automated, clinicians can focus on complex care.

However, KPIs must be balanced with safety metrics. Lower decision time is valuable only if clinical correctness and member protection remain intact. Continuous A/B testing, external validation, and safety monitoring should be embedded in product metrics.

Broader Industry Signals: Vendor Risk, Market Shifts, and High-Profile Incidents

The AI ecosystem is crowded and politically fraught. High-profile vendor dynamics provide context for any enterprise planning aggressive AI adoption.

• Regulatory and government scrutiny: Anthropic faced Pentagon questioning about the use of its models in classified systems, and allegations about data extraction by other vendors surfaced. Government procurement priorities and national-security concerns can reshape vendor access to certain markets and lead to stricter contractual requirements.
• Vendor conflicts and market responses: Meta’s multi-billion-dollar chip purchase deal with AMD — and similar vendor alignments — illustrate how large tech players secure hardware and influence model performance capabilities. Such supply dynamics affect cost and availability for enterprise participants.
• Reputational and operational fallout: IBM’s shares tumbled after Anthropic’s claims about a tool’s capacity to modernize Cobol sparked market confusion, prompting an IBM corporate response. The incident shows how model claims, even if technically framed, can create outsized market reactions.
• Service reliability controversies: Reports that Amazon Web Services outages involved internal AI coding tools drew a public rebuttal from Amazon that blamed misconfiguration. The episode highlights the operational risk when AI toolchains are integrated into production infrastructure.

Healthcare organizations must weigh vendor reliability, openness about model training data, security posture, and the capacity to supply verifiable safety documentation. Contractual protections around model performance, incident response, and indemnities will become standard procurement features.

Clinicians vs. Executives: The Trust Gap

Survey data shows a divergence in AI optimism between healthcare executives and clinicians. Executives largely see AI as a tool for quality improvement and staffing mitigation; clinicians remain cautious. That skepticism comes from a mix of reasons:

• Clinical risk: AI errors can directly harm patients; clinicians therefore require higher thresholds for model adoption.
• Workflow fit: Clinicians worry that AI will create new work—more documentation, explanation burdens, or incorrectly framed suggestions—rather than reduce load.
• Accountability and liability: When an AI system influences care decisions, clinicians face questions about responsibility and liability.

Earning clinician trust demands staged deployments with clear guardrails, explainable outputs, and integrated workflows that do not add cognitive load. Early wins in non-diagnostic settings—documentation automation, prior-authorization handling, administrative scheduling—can demonstrate value and build credibility.

Case study approach: a health system that pilots automated prior-authorization for routine imaging might show reduced turnaround times, fewer phone calls between providers and payers, and no change in diagnostic outcomes over an observation period. Those concrete outcomes provide clinicians evidence to expand automation to other domains.

Operationalizing Responsible AI: Governance Frameworks and Monitoring

Responsible AI requires more than policy memos. Elevance’s path suggests a concrete playbook that other payers and health systems can adapt:

1. Use case triage and risk classification

• Categorize AI projects by potential patient harm and data sensitivity.
• Low-risk internal assistants and member-facing informational bots get an accelerated path; clinical decision support and autonomous agents require rigorous clinical validation.

2. Data governance

• Enforce minimum necessary data principles.
• Maintain data lineage, versioning, and retention policies.
• Use synthetic or de-identified data for model development where feasible.

3. Model validation and monitoring

• Pre-deployment: run retrospective validation against labeled cases and clinician adjudication.
• Post-deployment: continuous monitoring for drift, error rates, and safety signals.
• Audit trails: store model versions, input-output pairs, and decision rationales.

4. Human oversight and escalation

• Define human-in-the-loop points for each workflow.
• Provide clear escalation paths and accountability matrices.
• Limit agent autonomy through explicit stop conditions and role-based permissions.

5. Vendor management

• Require vendors to disclose training data sources, robustness testing results, and red-team findings.
• Contractually mandate incident response timelines and security obligations.

6. Workforce education

• Train employees on responsible use, prompt engineering, and model limitations.
• Offer role-specific certifications and refreshers.

7. Transparency and member rights

• Provide members with clear notices when AI is used and how it affects their care or benefits.
• Maintain easy appeal and dispute resolution channels.

These components turn abstract governance into operational controls that protect patients, ensure compliance, and enable scalable adoption.

Balancing Speed and Safety: When to Automate and When to Wait

Healthcare leaders must decide where to move fast and where to be cautious. Prioritization criteria include risk of harm, potential for quick wins, data availability, and regulatory complexity.

• Automate when: processes are high-volume, routine, and low-clinical-risk—examples include claims routing, benefits FAQs, and documentation templates.
• Proceed cautiously when: decisions affect care access, diagnosis, or treatment; clinical validation data is limited; or legal/regulatory risk is high.
• Delay or design constrained pilots when: external vendor transparency is insufficient, privacy risks are unresolved, or clinician support is lacking.

Elevance’s strategy follows this gradient: member chat and internal assistants are reasonable early adopters; prior authorization automation is somewhere in the middle—high value but requiring careful oversight; agentic clinical workflows demand the most stringent controls.

Real-World Implications: Members, Providers, and Payers

Members: Conversational agents simplify plan navigation and help members find in-network providers, reducing surprise out-of-pocket costs. When done honestly—annotating uncertainty and providing next steps—these tools can reduce anxiety and accelerate access to care.

Providers: Automation of administrative work can reduce friction between practices and payers. Faster prior-authorization can improve cash flow for providers and reduce administrative overhead, mitigating a major pain point for ambulatory clinics.

Payers: Reduced out-of-network utilization, lower administrative costs, and better member engagement improve financial performance. AI-driven routing and recommendation engines can nudge members toward cost-effective, high-quality providers.

These benefits materialize only when automation is precise and trusted. False positives in prior-authorization automation or incorrect provider recommendations can erode trust quickly, increasing appeals and manual reviews.

External Market and Policy Forces That Will Shape Adoption

Technology choices do not occur in a vacuum. Several external forces will influence how health systems deploy AI:

• Government procurement and national security concerns: Military and classified systems scrutiny of AI vendors has market ripple effects. Vendors with opaque data practices may find enterprise contracts constrained or renegotiated.
• Hardware supply chains and cloud partnerships: Deals between hyperscalers, chipmakers, and large tech buyers shape performance economics and vendor leverage for enterprise adopters.
• Regulatory regimes and standards: Emerging standards for AI safety, model certification, and auditability will change procurement checklists. Organizations that align early will have competitive advantages.
• Public perception and litigation: High-profile model failures can trigger regulatory backlash and class-action litigation. Proactive transparency and auditability reduce exposure.

Enterprises will need legal, procurement, and compliance teams deeply integrated into AI projects to navigate evolving obligations.

Lessons from Incidents and Market Tensions

High-profile industry events provide cautionary lessons that should shape health-system strategies:

• Vendor claims about transformative capabilities can trigger market panic. The Cobol modernization episode shows how technical claims can have cascading commercial impacts even when the core value proposition is more limited. Enterprise buyers should require empirical validation and avoid adopting vendor narratives at face value.
• Incidents involving internal AI tools and outages illustrate that AI is an operational risk as much as a model risk. Proper configuration management, change control, and access permissions are essentials.
• Allegations of model-training data misuse underscore the need for contractual clarity on training-data provenance and non-derivative use clauses. Enterprises must demand vendor commitments about data handling.

These lessons reinforce the need for robust technical, contractual, and operational controls.

Practical Roadmap for Health Organizations Considering AI

Organizations preparing to adopt AI at scale can follow a tactical roadmap:

1. Inventory existing processes and data maturity.
2. Prioritize use cases by risk, value, and implementation feasibility.
3. Launch controlled pilots with measurable KPIs and safety endpoints.
4. Build governance: data controls, model validation routines, and monitoring.
5. Upskill the workforce with role-specific training and certification.
6. Negotiate vendor contracts that specify model transparency, audit rights, and incident responsibilities.
7. Scale iteratively, incorporating clinician feedback and real-world performance metrics.

This approach balances speed with prudence and provides a path to scale while preserving patient safety.

Where Elevance’s Strategy Could Lead the Industry

Elevance’s emphasis on member-facing conversational interfaces, prior-authorization automation, and enterprise assistants mirrors a broader industry trend of using AI to simplify administrative complexity. If Elevance successfully integrates narrow, well-governed agentic systems that alleviate administrative burdens while safeguarding clinical decisions, peer organizations will likely follow. The company’s scale—45.2 million medical members and substantial operating revenue—makes it both a testing ground and a potential exemplar.

Key indicators to watch:

• Changes in prior-authorization turnaround times and appeals rates.
• Member satisfaction related to plan navigation and cost transparency.
• Employee productivity gains attributable to internal assistants like Spark.
• Incidents or model failures and the speed of Elevance’s remediation.

Transparent reporting on these metrics would accelerate sector learning and help establish best practices.

The Limits of Technology: Why Human Judgment Remains Central

AI can accelerate processes, surface options, and reduce repetitive work, but it cannot replace human judgment in complex clinical decisions. Models make probabilistic inferences and can hallucinate or misinterpret subtle clinical nuance. The appropriate design is not AI that seeks to replace clinicians but AI that augments decision pathways and limits the cognitive load of administrative tasks. Elevance’s insistence on human oversight—even as humans oversee multiple agents—places people at the center of system design.

That human-centric stance aligns with clinician concerns and with ethical frameworks for responsible AI. It also reflects the pragmatic reality that patients and clinicians expect accountability and clear lines of responsibility when care decisions are made.

Final Considerations: Pace, Proof, and Public Trust

Enterprises must deliver proof points before expecting broad adoption. Executives’ optimism must be matched by clinician validation, measurable improvements in access and cost, and transparent governance that earns member trust. The interplay of vendor dynamics, public policy, and technical risk means organizations that invest in rigorous validation, workforce readiness, and robust governance will be better positioned to capture AI’s benefits without exposing patients to undue risk.

Elevance’s early deployments—Sydney, Spark, and prior-authorization automation—offer a practical template: prioritize high-volume, lower-clinical-risk workflows, measure outcomes, and build the human and technical scaffolding needed for more ambitious agentic deployments. If that blueprint proves effective at scale, it can become a model for the wider healthcare ecosystem.

FAQ

Q: What specific AI features has Elevance Health deployed for members? A: Elevance has integrated a conversational agent into its Sydney mobile app that answers natural-language questions about benefits and costs and recommends in-network providers. The system interprets plan details and geographic network data to deliver actionable guidance to members.

Q: How is Elevance using AI for prior authorization? A: AI automates portions of prior-authorization reviews by matching submitted documentation to coverage rules and clinical criteria. Low-risk or clear-cut cases can be processed automatically, while ambiguous or high-stakes cases are escalated to human reviewers. The approach prioritizes speed and reduced administrative burden while retaining clinician oversight for complex decisions.

Q: What is Spark and how widely is it used? A: Spark is Elevance’s internally developed enterprise chatbot that leverages multiple large language models to assist associates with document analysis and common tasks. More than 60,000 Elevance employees used Spark, producing roughly 10 million messages in the first half of 2025.

Q: What is agentic AI and why is Elevance exploring it? A: Agentic AI refers to systems that can perform multi-step tasks autonomously by composing actions and API calls. Elevance sees agentic systems as promising for orchestrating complex benefits and claims workflows. The company plans to scale agentic use cases more aggressively in 2026 but is establishing strict task boundaries and supervision protocols before broad deployment.

Q: How does Elevance ensure AI safety and compliance? A: Elevance focuses on training agents only on task-relevant data, maintaining human oversight, deploying governance frameworks for data and model monitoring, and upskilling employees through AI fluency certifications that cover responsible AI, prompt engineering, and advanced workflows. The company also uses a mix of internally developed models and enterprise licenses with external vendors to balance control and capability.

Q: What are the main risks associated with AI in healthcare? A: Primary risks include incorrect or unsafe clinical advice, data privacy breaches, biased outcomes from imperfect training data, operational failures in integrated toolchains, and vendor opacity regarding model training. Reputational and regulatory consequences can follow high-profile failures.

Q: How do clinicians view AI compared with executives? A: Executives tend to be more optimistic—three out of four believe AI can positively impact care quality. Clinicians are more skeptical: only 45% believe AI can address staffing shortages to some extent, and just 6% think AI can solve the problem mostly or completely. Building clinician trust requires validated, integrated, low-risk deployments that demonstrably reduce workload.

Q: What should other health organizations learn from Elevance’s approach? A: Adopt a staged strategy: prioritize low-risk, high-impact automation; enforce data-minimization and auditing; build human-in-the-loop controls; train staff in AI fluency; and require vendor transparency. Use concrete performance metrics and safety endpoints to justify expansion into higher-risk clinical domains.

Q: Will agentic AI replace human jobs in healthcare? A: Agentic AI may change the nature of some roles by automating repetitive administrative tasks, which can reduce workload and reallocate staff toward higher-value activities. Elevance envisions humans overseeing multiple agents rather than eliminating decision-makers. Upskilling programs are intended to help employees transition to supervising and validating AI outputs.

Q: How should members be informed about AI use? A: Members should receive clear disclosures when AI affects their care or benefits interactions, including how AI is used, when a human will review decisions, and how to appeal automated determinations. Transparency strengthens trust and aligns with evolving regulatory expectations.