Enforcing CMS-0057-F decision timeframes on prior authorization AI agents
CMS-0057-F sets a 72-hour clock for expedited and a 7-calendar-day clock for standard prior authorization decisions, and requires a specific reason for denials. A runtime policy layer in the AI execution path can enforce parts of this at the level of individual interactions and tool calls: escalate to a human reviewer as a deadline approaches, and block a denial communication that lacks a specific reason. It cannot own the payer's overall decision clock, queue management outside the AI path, or the clinical decision.
Evidence to retain for a regulator or auditor
- Request-received, status-change and decision timestamps linked to each policy decision on an AI action.
- The policy version in force at the time of each decision.
- The action taken on each interaction or tool call: allow, redact, block, rate limit or escalate.
- Escalation records, including trigger, time raised and assigned reviewer.
- Human approval actions with who, when and outcome.
- Blocked denial attempts and the missing-reason condition that caused the block.
What the rule requires, and for whom
CMS-0057-F (89 FR 8758) is a final rule, effective April 8, 2024. It requires certain payers to decide prior authorization requests within 7 calendar days for standard requests and 72 hours for expedited requests. The finalized timeframes apply to Medicare Advantage organizations, state Medicaid and CHIP fee-for-service programs, Medicaid managed care plans and CHIP managed care entities, for non-drug items and services. An extension of up to 14 days is possible in certain circumstances.
Scope matters before any policy is written. Qualified health plan issuers on the federally-facilitated exchanges were not given these timeframes. Their existing requirement is no later than 15 days for standard decisions and no later than 72 hours for expedited requests. Drugs are outside the finalized timeframes. An April 2026 proposed rule would add drug decision timeframes, with proposed compliance dates beginning October 1, 2027. As of November 2026 it is proposed only and may change.
The rule also requires a specific reason when a request is denied, communicated through the Patient Access API. For state CHIP fee-for-service programs, 42 CFR 457.732 requires a specific denial reason in the response to the provider, excluding drugs, beginning January 1, 2026. Separately, Prior Authorization API compliance dates fall in 2027. Payers must also make prior authorization information available through the Patient Access API within one business day of receiving a request and update it within one business day of any status change.
Which obligations map to runtime enforcement
Runtime policy acts on what passes through it: an agent interaction or a tool call. That makes some obligations enforceable and others only observable. The mapping below is a practitioner's reading of the rule, not CMS guidance, and the thresholds are internal design choices.
- Approaching a decision deadline: Enforceable as escalation. When elapsed time against the 72-hour or 7-day clock crosses an internal threshold, route the case to human approval rather than letting the agent continue.
- Denial without a specific reason: Enforceable as a block. Treat a specific reason as a precondition for any AI-generated denial communication, and stop the tool call that would send it if the reason is absent.
- Stalled or repeated tool calls: Partly enforceable. Rate limits and timeout conditions can contain retry loops, and escalation can follow, but only for calls the layer sees.
- Sensitive data in tool calls: Enforceable as redact or block, limiting what each tool call receives to what the step requires.
- The payer's overall decision clock: Not enforceable at this scope. It depends on upstream intake, queue and case-management systems.
- Clinical determination and API implementation: Not enforceable at this scope. Medical necessity decisions and Prior Authorization API build-out sit with clinical staff and engineering programs.
Translating the clocks into policy
Start by separating three things that are easy to conflate: the regulatory deadline, the internal escalation trigger, and the policy action. The regulatory deadline is fixed by the rule for each line of business. The escalation trigger is yours to set, and it should fire well ahead of the deadline so a human reviewer has time to act. The policy action is what the runtime layer does when the trigger fires.
Policy conditions need inputs. Each clock is per request, so the layer needs a request-received timestamp and an expedited or standard classification. Those originate in upstream systems. Confirm that the layer can read both at every tool call, and decide what it does when either is missing. A sensible default is to escalate rather than proceed.
Do not encode extension eligibility into default policy. The 14-day extension applies only in certain circumstances, so treat it as a separate rule that compliance counsel confirms before it influences any automated path.
- Map each line of business to its applicable timeframe: 7 days and 72 hours, or the 15-day and 72-hour rule for exchange QHP issuers.
- Check state and program-specific timeframes that may be shorter, and set the stricter clock as the policy input.
- Define internal escalation thresholds per request type, kept distinct from the regulatory deadline.
- Require a specific denial reason before any denial communication is allowed.
- Define the fallback when timestamp or classification data is missing.
- Name the human approver role for each escalation and the expected response window.
Limits of enforcement at the interaction and tool-call level
An enforcement layer can only act on interactions that pass through it. A request routed around the AI path, or stalled in a queue the layer does not observe, is outside its reach, and a missed deadline there will not show up as a policy event. Compliance leaders should state this boundary plainly in internal documentation, so runtime evidence is not mistaken for the payer's full compliance record.
This is also why runtime policy complements rather than replaces process controls. Queue monitoring, intake service levels and public reporting remain with upstream systems and operations teams. Runtime enforcement narrows one specific gap: the AI agent's own behavior at each step, including whether it escalated when it should have and whether it was permitted to issue a denial without a reason.
The practical tradeoff is escalation volume. Early thresholds produce more human reviews and protect the deadline. Late thresholds reduce reviewer load but leave less time to act. Tune thresholds against reviewer capacity, and revisit them as the Prior Authorization API requirements take effect in 2027.
Where Trussed fits
Trussed enforces policy in the execution path across models, agents, tools and MCP servers, with actions to allow, redact, block, rate limit and escalate for human approval. The scope is individual interactions and tool calls. Policy is enforced at runtime, not reviewed after the fact, and the enforcement generates audit-ready evidence. Deployment is self-managed in the customer cloud, with configurable retention and data residency. Trussed does not provide transaction-level governance or end-to-end tracing of agentic workflows.
Trussed has run continuously in production at a regulated enterprise for more than 18 months, and serves customers in insurance, pharma, healthcare and education. The conditions described above, such as elapsed-time thresholds and denial-reason checks, are policies a payer defines. Whether a given condition can be evaluated depends on the timestamp and classification data your upstream systems supply.
From regulatory clock to runtime policy
Requirement
72-hour expedited and 7-calendar-day standard decisions; specific denial reason.
Policy condition
Elapsed time, request type and denial-reason presence evaluated on each interaction and tool call.
Runtime action
Allow, redact, block, rate limit, or escalate for human approval.
Evidence
Timestamped policy decisions, escalations and approvals tied to the policy version in force.
Review how your timeframe policy would be enforced
Walk through your lines of business, escalation thresholds and evidence requirements with a specialist.
Talk to an ExpertHow does your AI governance program compare?
See where your program has gaps in less than 2 minutes.