Diagnosing Danger

The legal risks of AI-generated diagnoses in Medicare Advantage coding.

The government is going after companies and employees for allegedly adding diagnoses that fraudulently increase payments to Medicare Advantage (MA) plans. This has become more prevalent with the use of artificial intelligence (AI) programs that scan documentation and suggest potentially missed diagnosis codes. The legal implications of this practice can be serious and far-reaching, especially when the result is higher MA risk adjustment payments.

False Claims Act Liability

AI-assisted coding that leads to the submission of inaccurate or unsupported diagnoses is a False Claims Act (FCA) violation. Under the FCA, both organizations and individuals (including coders, auditors, and possibly AI vendors) can be held liable for submitting or causing the submission of false claims to the government for reimbursement.

Example: If AI suggests a diagnosis not clearly supported by clinical evidence, and it is submitted without human validation, this could constitute a false claim.

Corporate Responsibility and Oversight

Organizations using AI are responsible for ensuring that the output of AI-generative tools is: 

• Clinically valid,
• Auditable, and
• Compliant with the Centers for Medicare & Medicaid Services’ (CMS’) guidelines.

Failure to implement robust governance over AI tools may be interpreted as negligence or willful ignorance.

Human Accountability

Even when AI tools are used, human reviewers remain accountable for final code selection. If employees rely solely on AI suggestions without validating against MEAT (monitor, evaluate, assess/address, treat) criteria, they could be implicated for negligent or fraudulent coding.

AI can help when:

• Used as an assistive tool (e.g., highlighting potential diagnoses, flagging missing MEAT elements)
• Integrated into a workflow with human oversight
• Trained with clinical context and updated regulatory standards

AI can make matters worse when:

• It autonomously adds codes without review.
• Organizations prioritize production over accuracy.
• There is overreliance on AI output in lieu of clinician documentation.

Proactive Steps to Compliance

Take measures to ensure ethical use of AI in coding. For example:

• Human-in-the-loop validation: All AI-suggested codes must be reviewed and validated by a certified coder.
• Audits: Perform regular audits of AI-generated output for compliance with CMS guidelines.
• Documentation transparency: Ensure that coding decisions (whether by AI or a human) are supported by clearly documented clinical evidence.
• Policy development: Establish internal policies governing the use of AI in risk adjustment, including accountability, version control, and update cycles.

AI has the potential to either resolve or worsen coding — it depends entirely on how it’s implemented.

AI is useful when:

• It’s used ethically as a support tool, not a decision-maker.
• Coders are trained to validate AI-suggested codes using MEAT criteria.
• Organizations invest in transparency, audit trails, and compliance reviews of AI output.
• It’s used to identify under-documented or missed diagnoses, improving coding completeness and accuracy without inflating risk.

Example: AI can flag incomplete documentation or highlight clinically relevant information that a human may overlook, helping prevent undercoding and ensuring fair reimbursement.

AI is not useful when:

• It’s used to game the system by maximizing risk scores without medical necessity.
• There’s no human oversight, or coders are pressured to accept AI suggestions uncritically.
• Vendors and providers use AI to auto-code unsupported diagnoses in pursuit of higher payments.
• It contributes to “black box” coding, where coders can’t explain why a diagnosis was chosen, undermining audit defensibility.

Example: If AI pulls a chronic condition from a past encounter and applies it to the current year without new documentation, that could lead to unsupported hierarchical condition category (HCC) capture. AI doesn’t distinguish between active and historical conditions unless explicitly programmed to do so.

Ensure Compliance

Practices can proactively ensure AI is used ethically in coding by implementing a governance framework that blends compliance, transparency, and accountability. Here’s a structured approach:

1. Establish clear policies for AI use.

• Define the role of AI as assistive only, not autonomous.
• Require human review and final validation for all codes suggested by AI.
• Prohibit auto-submission of AI-generated codes without coder intervention.

2. Implement AI-specific compliance audits.

• Regularly audit coded charts for MEAT and clinical validity.
• Monitor for patterns of overcoding or unsupported diagnoses.
• Compare AI output with human coder findings to assess accuracy and bias.

3. Ensure transparency and explainability.

• Use AI systems that offer explanations or traceability for their suggestions.
• Require vendors to disclose algorithms, update history, and training data sources.
• Document who reviewed and approved each code to maintain an audit trail.

4. Invest in coder training and oversight.

• Train coders on how to evaluate and challenge AI output.
• Educate teams about AI limitations, ethical concerns, and legal risks.
• Encourage coders to report any inconsistencies or AI “hallucinations.”

5. Include compliance and legal concerns in vendor selection.

• Vet vendors for tech performance as well as for regulatory awareness.
• Require vendors to provide risk assessments and validation studies.
• Include clauses in contracts and hold vendors accountable for errors or misuse.

6. Foster a culture of ethical awareness.

• Promote coding accuracy over production volume.
• Recognize and reward ethical practices, such as flagging unsupported AI suggestions.
• Make ethical coding part of ongoing education, not just a one-time training.

The legal consequences of submitting invalid codes are FCA violations, CMS sanctions, and reputational damage.

Risky Business

Without human oversight, transparent AI logic, and real-time auditing, AI is a danger to your organization. Technology does not eliminate risk — it magnifies it when ethics and governance are missing.

• Author
• Recent Posts

Dannilla Morgan

Dannilla Morgan, CPC, CBCS, is a medical coder specializing in risk adjustment and healthcare compliance, with more than four years of experience in the industry. She serves as the president of the Carmel, New York local chapter and previously served as member development officer.

Latest posts by Dannilla Morgan (see all)

• Diagnosing Danger - November 1, 2025
• Accuracy vs. Production in Risk Adjustment Coding - September 2, 2025