What is AI accountability?

By Verillian. Published . Updated .

AI accountability means a named person answers for how AI is used and can show what happened. Who holds it, what standards ask for and the record it takes.

A woman in a blazer opens a beige folder at a desk with stacked files, a calculator and a laptop.

AI accountability means a named person answers for how AI is used at your organization, and can show what happened when someone asks. The software can’t carry that for you. The duty stays with your people and your agency, and meeting it takes an owner, a policy and a record you hold.

What does AI accountability mean?

The federal government’s AI Accountability Policy Report (NTIA, March 2024) describes it as “a chain of inputs linked to consequences”. Information about an AI system supports evaluation, and evaluation feeds into consequences such as liability and regulation. That report is aimed at government policy.

Inside an organization that uses AI, the same chain comes down to who owns each use, what your policy allowed, and what happened. The first two take a decision and a document. The third takes evidence, and a policy can’t supply it.

Who is accountable when AI gets it wrong?

The people and the organization that used it. Oklahoma’s standard for state agencies puts it in one sentence. “Use of an AI Agent or Agentic System does not transfer responsibility from a state agency, employee, contractor, supplier or public official to the AI Agent or Agentic System.”

Guidance for Washington’s local governments lands in the same place for everyday tools, in StateTech’s report of it. “Human accountability remains essential, even when AI assists with drafting or analysis.”

Courts are already holding people to it. In Johnson v. Dunn (N.D. Ala. 2025), lawyers whose client was an Alabama government agency filed case citations made up by generative AI, and the court said the conduct “demands substantially greater accountability than the reprimands and modest fines that have become common as courts confront this form of AI misuse.”

What do standards and regulators ask for?

They start with a named owner. NIST’s AI Risk Management Framework, which is voluntary, asks for accountability structures so that “the appropriate teams and individuals are empowered, responsible, and trained.” OMB memo M-25-21 says federal agencies “must ensure human oversight, intervention, and accountability suitable for high-impact use cases.”

Then three of them name the log:

  • New York’s guideline for state agencies says “All interactions with approved GenAI tools (e.g., ChatGPT, Copilot) should be logged for a minimum of 90 days to enable accountability and help identify potential misuse, security issues, or areas for improvement in system usage.”
  • Oklahoma, for AI agents, says “logs sufficient to support oversight, auditing, investigation, reconstruction, replay of consequential actions and incident response.”
  • FINRA’s 2026 oversight report, which is guidance and not a rule, says monitoring “may include storing prompt and output logs for accountability and troubleshooting.”

Most policies we’ve read haven’t got there yet. Of the 65 government AI policies we could read, ten name prompts as something to keep.

What does it take in practice?

  1. Name an owner for each AI tool and use. Oklahoma goes further for agents. If “accountable ownership becomes unclear,” the agent “shall be suspended or deactivated until sponsorship is formally restored.”
  2. Write down what’s allowed, and apply it where the work happens. A written policy on its own is a statement of intent.
  3. Keep a record you hold, showing who sent what to which AI tool, what came back and what your policy decided. Our post on AI audit trails sets out what that record needs. An auditor’s version of the question is in AI audit evidence for a state agency.
  4. Give someone the job of reading it, on a schedule you can show.

What record does Verillian keep?

Verillian keeps a signed record of every captured AI interaction on your own servers, so the person who answers for AI use can show who sent what, to which AI tool, and what your policy decided.

Sources

  • NTIA, AI Accountability Policy Report (March 2024). Report.
  • Oklahoma, standard for state agencies on AI agents and agentic systems. Standard text.
  • StateTech, report on guidance for Washington’s local governments. StateTech report.
  • Johnson v. Dunn (N.D. Ala. 2025). Opinion text.
  • NIST, AI Risk Management Framework. Framework text.
  • OMB memo M-25-21. Memo text.
  • New York State, guideline for state agencies on generative AI. Guideline text.
  • FINRA, 2026 oversight report. Report text.

Sources read on 4 November 2026.

What the record keeps

See Verillian govern AI on a real device.