The Great Friction

AI Accountability Map · Symposium poster companion

When AI causes harm, who is accountable?

Seven questions help trace responsibility from an AI-influenced decision to its human consequences, the evidence behind it, and a path to remedy.

Developed by Rosa J. Combs, J.D. · Founder & Managing Director, The Great Friction

Download the full map Explore the questions

Free PDF · Seven questions and a printable worksheet

The method

Follow the decision all the way through.

Use these questions in order, then return to any gap where an answer is missing or contested.

1

The Decision

What decision did the AI system influence?

2

The Impact

Who was affected, and what harm occurred?

3

The Actors

Who designed, purchased, configured, deployed, monitored, and relied on the system?

4

The Evidence

What records demonstrate how the decision was made?

5

Human Oversight

Where did a human enter the process, and did that person have meaningful authority?

6

The Accountability Gap

Where can responsibility be denied, transferred, or obscured?

7

Recourse

How can the affected person challenge the decision, get an explanation, and obtain a remedy?

The poster case

The map in practice: post-acute care decisions

The nH Predict litigation raises questions about how a prediction may influence coverage and how responsibility is shared among a tool maker, payer, reviewers, and clinicians.

About the record. Patients allege that nH Predict was used in coverage decisions. UnitedHealthcare disputes that characterization. The court has allowed contract claims to proceed and ordered discovery about the tool’s design, use, and oversight. The questions below are an analytical exercise, not findings on the disputed facts.

A separate Senate investigation reported rising UnitedHealthcare post-acute prior authorization denial rates during a period of increasing automation; it did not resolve this lawsuit’s disputed facts.

  1. 1DecisionLength of stay and coverage for post-acute care.
  2. 2ImpactPatients facing a possible earlier discharge, denial, or appeal.
  3. 3ActorsnaviHealth, UnitedHealthcare, care coordinators, reviewers, treating clinicians, and patients.
  4. 4EvidenceExamine model documentation, clinical records, decision records, notices, and appeals.
  5. 5Human oversightWho reviewed a prediction, and could that person meaningfully depart from it?
  6. 6Accountability gapWhere could responsibility shift among vendor, payer, reviewer, policy, and algorithm?
  7. 7RecourseHow could a patient contest the decision and obtain a clear explanation or remedy?

Read the 2025 court order, the 2026 discovery order, and the Senate investigation summary.

Source material

Governance resources

The map is designed to help practitioners ask questions alongside established frameworks and guidance. These sources provide the original requirements and context; the map itself is an independent teaching and analysis tool.

NIST AI Risk Management FrameworkVoluntary framework for managing risks across the AI lifecycle.IEEE 7000-2021Systems design process for addressing ethical concerns.EU AI ActOfficial text of Regulation (EU) 2024/1689.FDA digital health guidanceClinical decision support and AI-enabled device guidance.

Continue the conversation

Apply the map to a decision your organization faces.

Bring The Great Friction to your team for a case-based presentation, workshop, or research collaboration.

Discuss a collaboration Rosa@thegreatfriction.com