Point of View
Professional Services · AI Governance · Risk Management
Who Owns the Decision When AI Is Wrong
When a major professional services firm delivered a $440,000 government report full of fabricated citations and a fake court quote, the firm refunded part of the fee and stood by its conclusions. No one was named as responsible. The incident wasn't really about AI making mistakes -- it was about no one having decided, in advance, who owns an AI-assisted decision when it goes wrong.
Apr 2025
Points of View is Vaxa's opinion column. Each article states a position on a question where reasonable people disagree — and makes the argument for that position directly.
VAXA POINT OF VIEW
When a major professional services firm delivered a $440,000 government report containing fabricated academic citations and a fake quote attributed to a federal court judge, the firm partially refunded the fee and stood by the report's conclusions. No individual was named as responsible. No one was reported terminated. The incident wasn't really a story about AI making mistakes — it was a story about no organization having decided, in advance, who owns an AI-assisted decision when it turns out to be wrong.
Pattern
A recent, well-documented case makes the shape of the problem clear. A Big Four consulting firm was commissioned by a national government to produce a 237-page review of a welfare compliance system, for a fee north of $400,000. An outside academic identified roughly twenty fabricated references in the delivered report — citations to papers that didn't exist, attributed to real researchers who'd never written them — plus a quote falsely attributed to a federal court judge. The firm eventually confirmed generative AI had been used in preparing the document. It issued a corrected version, agreed to a partial refund, and publicly maintained that the report's substance and recommendations were unaffected. No individual consultant, partner, or reviewer was named. No public disciplinary action was reported. The firm treated it as a billing correction, not an accountability event.
That response wasn't unusual — it was rational, given the current landscape. There is no established norm across professional services for what happens when an AI-assisted deliverable turns out to contain fabricated content. Absent an explicit answer, organizations default to the answer that costs the least: a partial refund, a quiet correction, a defense of "substance."
Silence
A recent, well-documented case makes the shape of the problem clear. A Big Four consulting firm was commissioned by a national government to produce a 237-page review of a welfare compliance system, for a fee north of $400,000. An outside academic identified roughly twenty fabricated references in the delivered report — citations to papers that didn't exist, attributed to real researchers who'd never written them — plus a quote falsely attributed to a federal court judge. The firm eventually confirmed generative AI had been used in preparing the document. It issued a corrected version, agreed to a partial refund, and publicly maintained that the report's substance and recommendations were unaffected. No individual consultant, partner, or reviewer was named. No public disciplinary action was reported. The firm treated it as a billing correction, not an accountability event.
That response wasn't unusual — it was rational, given the current landscape. There is no established norm across professional services for what happens when an AI-assisted deliverable turns out to contain fabricated content. Absent an explicit answer, organizations default to the answer that costs the least: a partial refund, a quiet correction, a defense of "substance."
The firm had an AI usage policy. It did not, apparently, have an AI accountability policy — and only one of those two things determines what happens when something goes wrong.
Precedent
The professions that already solved this problem offer a genuine template, not just a cautionary contrast. A doctor who relies on a diagnostic algorithm is still the one who signed the chart. A lawyer who uses AI for legal research is still the one who filed the brief, and several have already faced sanctions for AI-fabricated case citations precisely because that individual accountability already exists in the legal profession — several U.S. judges have sanctioned attorneys directly, by name, for submitting briefs with AI-hallucinated case law. The mechanism worked in law specifically because the profession already had a named, personal point of accountability to attach the failure to.
Professional services broadly does not have that mechanism for AI-generated deliverables the way law and medicine do for their core practice. A report is delivered under the firm's name and the client relationship's authority, not under one individual's personal professional license. That structural difference is exactly why the consequence lands on the firm's balance sheet instead of an individual's career — and why, absent a deliberate policy choice, it will keep landing there.
Implication
The organizations at risk here aren't the ones using AI in client deliverables — nearly all of them are, at this point. The organizations at risk are the ones that haven't yet answered, in writing, before an incident happens: who reviews AI-assisted output before it goes out under the firm's name, what the verification standard is, and what the actual consequence is when that standard isn't met. Firms that wait until an incident to improvise an answer will keep landing on the same one the public case above did — a refund, a defense of substance, and no named accountability — because that's the default answer when no other one has been decided in advance.
The Question to Ask
If your organization delivered a client-facing report tomorrow that turned out to contain fabricated AI-generated content, do you already know who would be accountable for it — or would that decision get made for the first time in the moment it happened?