Bill
A 11048
Requires AI in utilization review to be transparent and overseen by clinicians; AI cannot be sole basis for denials and must disclose criteria, data, and outcomes.
Purpose and intent
- This bill adds explicit requirements and oversight to the use of artificial intelligence–based algorithms in the utilization review (UR) process used by health insurers, to ensure transparency, accountability, and patient safety.
- It aims to prevent AI from being the sole basis for denial of care, mandate clinician involvement, and provide patients and providers with information about AI usage and criteria.
Key provisions and changes
- Definitions (Section 1)
- Introduces terms such as “Artificial intelligence-based algorithm,” “Utilization review,” “Clinical peer reviewer,” and clarifies related meanings under the Insurance Law.
- Notice and transparency (Section 1(b))
- Insurers must provide written notice to insureds, enrollees, and health care providers about the use of AI in UR.
- Notice timing: when AI-based algorithms are first adopted and at least once per policy period; clear, conspicuous notices on insurer websites.
- If AI is used, insurers must disclose to the insured/enrollee’s provider (and to the insured/enrollee upon request):
- Criteria governing the AI algorithm
- Training data sets
- The AI algorithm itself
- Outcomes produced by software using the AI
- Adverse determinations must disclose AI involvement in notices to insureds/enrollees and providers, under existing adverse-determination procedures.
- Prohibitions: AI cannot base adverse determinations solely on group data; must not cause harm; must be auditable; data used only for intended purpose and in compliance with HIPAA.
- Submissions and standards (Section 1(c))
- Insurers must submit AI algorithms and training data sets to the Superintendent, with certifications that:
- Bias minimization is addressed (race, color, religion, ancestry, age, sex, gender, national origin, disability) and in line with anti-discrimination laws
- Adherence to evidence-based clinical guidelines
- Do not rely on non-compliant information for UR
- Do not independently create or alter clinical standards or coverage criteria
- Use of AI in UR determinations (Section 1(d))
- If an insurer’s UR process initially uses AI, adverse determinations must ultimately be made by clinical peer reviewers.
- Clinicians must consider provider recommendations, patient histories, and individual circumstances; must document and follow applicable UR laws.
- For emergencies, a prudent layperson standard applies.
- Insurers must report data on time spent by clinical peer reviewers before issuing an adverse determination.
- AI cannot be the sole basis for denial, delay, or modification of services based on medical necessity.
- Clinical peer reviewers involved in adverse determinations must sign a statement complying with professional-review requirements.
- Quality assurance (Section 1(e))
- Insurers must implement ongoing QA testing for AI algorithms with defined safety/efficacy parameters.
- Results must be submitted to the Superintendent and publicly posted within 30 days of submission.
- Penalties (Section 1(f))
- Violations may incur penalties including license suspension/revocation, caps on licenses issued, and monetary fines:
- Insurer violations: up to $5,000 per violation; up to $10,000 for willful violations; aggregate annual cap of $500,000.
- Clinical peer reviewer violations: up to $5,000 per violation; up to $10,000 for willful; aggregate annual cap of $100,000.
- Penalties are in addition to other remedies.
- Rulemaking (Section 1(g))
- The Superintendent is authorized to promulgate rules to implement the act.
Impact and who is affected
- Affected entities:
- Insurers (including accident/health insurers, certain corporations under Article 43, and HMOs under Public Health Law Article 44)
- Clinical peer reviewers and other UR professionals
- Health care providers and insured/enrollees who undergo utilization reviews
- Practical effects:
- Increased transparency around AI usage in UR (criteria, training data, and algorithm specifics)
- Clinician oversight and requirement that AI be a supporting tool rather than the sole determinant in most adverse decisions
- Public reporting of AI QA results and enhanced annual disclosures
- Potential administrative burden for insurers to compile and submit AI systems and training data, and to maintain documentation for audits
- New enforcement mechanism with monetary penalties and potential license actions for noncompliance
Effective date and implementation
- Effective 60 days after enactment.
- Immediate authority to adopt necessary rules and regulations to implement the act’s provisions.
Notes
- The bill includes harmonized definitions of “emergency condition” across multiple sections of the Insurance Law and Public Health Law, aligning with existing UR standards and ensuring consistency in emergency determinations.
- It emphasizes patient protections, data privacy, and anti-discrimination safeguards in AI-enabled utilization review.
Compiled from official sources — confirm details with the bill’s official record.
Sign in to ask a question.
