Skip to content
← All industries

Healthcare and health plans

Back-office operations with a high volume of documents and explicit rules on the autonomy of AI agents — CFM Resolution 2.454/2026 treats the degree of autonomy as a risk-classification factor.

13%of Brazilian hospitals formally adopt AI (18% of healthcare facilities use it)Cetic.br

Declared scope limit

Scope deliberately restricted to the back office: billing, claim denials, authorization, and medical bill auditing. AI agents never make clinical decisions. CFM Resolution 2.454/2026 is explicit, and we follow it strictly.

Agent scope: back office

  • Billing
  • Claim denials
  • Authorizations
  • Medical bill auditing
CFM Resolution 2.454/2026

Out of scope: clinical decisions

Diagnosis, prognosis, and treatment are decisions made by people.

Processes fit for AI agents

  • Medical bill auditing and claim denials
  • Billing and verification
  • Procedure authorization
  • Health plan document checking

Examples of units of work

  • Medical bill audited
  • Document checked
  • Request processed

Recommended Skill Pack

Medical bill auditing, approvals, and claim denials

  1. Medical bill received

  2. Check against the health plan contract and fee table

  3. Audit the bill and propose a disallowance

  4. Disallowance or denial: an identified auditor decides

    An identified person decides

    LGPD, art. 20
  5. Medical bill audited

    Billable unit

Evidence

A reference flow, not your company's process.

When the other side has an agent

Procedure authorization is already a negotiation between two sides: the provider requests, the health plan authorizes. With agents on both sides, each one needs an owner and a record, and the clinical decision stays out of scope.

Compliance

Compliance standard examples

Res. CFM 2.454/2026 (Federal Council of Medicine)
Prohibits delegating the final diagnostic, prognostic, or therapeutic decision to AI systems. Requires clinical record keeping and continuous auditing, and treats the degree of autonomy as a risk-classification factor.
LGPD, art. 20 (Brazil's data protection law)
Health data is sensitive data; review of an automated decision is the data subject's right.

Our read

Formal adoption of AI in healthcare remains low, but the opportunities are huge, making the document back office the lowest-risk entry point and the one with the highest immediate return.