The CPS’s admission that it fabricated legal authorities demonstrates institutional breakdown in one of Britain’s most critical prosecutorial agencies
The Crown Prosecution Service (CPS) has admitted to presenting two completely fabricated legal authorities to the High Court during extradition proceedings against Romania. The citations did not exist. They were generated by artificial intelligence. The revelation, which emerged in July 2026, exposes a fundamental failure in institutional oversight — not just of technology deployment, but of the most basic quality controls that should govern criminal prosecution.
What This Case Actually Reveals
The error was not a one-off human mistake caught by competent internal review. It was a systemic failure: a junior counsel used artificial intelligence to research legal authorities without verification, submitted false citations to court, repeated them in formal written submissions, and the agency’s own internal quality assurance failed to catch it. The judge’s written comment is direct: “While the increasing use of artificial intelligence in legal work may be both necessary and beneficial, the episode highlights the risks of its use without appropriate oversight.”
What this case reveals is not merely a technology problem. It exposes an institutional culture in which a critical agency — responsible for prosecuting serious crimes on behalf of the state — has adopted powerful new tools without implementing basic safeguards. The CPS knew AI systems hallucinate. It failed to require verification. It failed to catch the error internally. It had to be corrected by defense counsel.
How the Fabrication Reached the Court
The incident occurred during High Court proceedings in extradition appeals to Romania. The CPS, acting as respondent, filed written submissions that cited two non-existent legal authorities. These citations appeared first in the grounds of opposition prepared by the CPS, then were repeated verbatim in further documents prepared by junior counsel.
The judge did not identify the false citations. The CPS did not catch them in internal review. The appellants’ counsel — the defense — identified the fabricated cases and brought them to the Court’s attention. Only then did the CPS admit the error.
In its letter dated 3 March 2026, the CPS accepted that the citations were inaccurate and stated that they “likely originated from the use of artificial intelligence.” The agency later issued a second apology, admitting that its own junior counsel had independently identified the fake cases just before the appellants’ counsel did — meaning the error was discovered through external pressure, not internal competence.
AI Hallucinations: Known Risk, Unmanaged Failure
AI systems — specifically large language models used for legal research — are known to produce “hallucinations”: false but plausible-sounding citations, case names, and legal authorities that never existed. This is not a novel or surprising failure. Legal researchers have documented this risk repeatedly since 2023. Bar associations have issued guidance. Legal ethics committees have published warnings.
The CPS, as a large institutional actor with substantial resources, cannot credibly claim ignorance. Yet the agency’s response makes clear it had not implemented basic safeguards:
- No mandatory verification protocol — The junior counsel used AI-generated legal research without verifying the citations before formal submission to court.
- No internal QA catch — The CPS’s own review process failed to identify false authorities in documents filed on behalf of the state.
- No technical controls — The agency had not implemented systems to flag AI-generated content or require secondary verification.
The judge’s written finding was measured but direct: The operative cause was “human error in the failure to verify the authorities relied upon in formal submissions placed before the court.” But the institutional failure preceded that human choice — the CPS created conditions where unverified AI output was acceptable practice.
78 Cases Under Review — And What That Doesn’t Mean
Following the incident, the CPS conducted an internal review of 78 cases involving the same junior counsel. According to the agency’s own statement to the court, no similar issues were identified in that review, and the lawyer’s work was otherwise assessed as meeting standards.
This finding creates two distinct problems. First, it may indicate a real systemic issue that the review failed to detect. A single AI hallucination caught in one case does not mean other cases are error-free — particularly if the underlying verification protocols were deficient. The absence of similar errors identified in 78 cases does not prove similar errors don’t exist in those files; it may mean the review process was insufficient to detect them.
Second, it reveals inadequate institutional accountability. Even if this instance was truly isolated, the CPS’s response amounts to spot-checking rather than systemic reform. A competent institutional response would include: mandatory retraining, new verification protocols for all AI-assisted research, audit of all recent filings using similar AI tools, and public commitments to procedural changes. The CPS has not publicly announced any such measures.
Why This Matters: Prosecutorial Duty and Institutional Failure
The CPS occupies a unique position in the British criminal justice system. It is not a court. It is not a defense counsel. It is the agency authorized to prosecute serious crime on behalf of the Crown. Its failures are not merely professional mistakes — they are institutional failures with implications for the fairness and integrity of prosecutions themselves.
Under the Criminal Procedure and Investigations Act 1996, and established case law, prosecutors owe a duty to the court to ensure the accuracy and integrity of information submitted in criminal proceedings. This duty exists precisely because prosecutors wield state power and courts must be able to assume the Crown’s submissions are competent and honest.
The CPS’s submission of false legal authorities — whether generated by AI or human error — violates the core principle of that duty: that prosecutorial submissions have been competently reviewed and verified. The fact that the error was generated by AI does not diminish the institutional responsibility; it enlarges it. The CPS chose to deploy AI. The CPS failed to implement safeguards. The CPS is therefore responsible for the consequences.
The Broader Pattern: Institutional Competence in Decline
This incident is not isolated. Across 2026, British institutions have experienced a cumulative erosion of public confidence. A Legal Aid Agency data breach exposed 18 years of applicant information. An HMRC phishing attack cost £47 million. UK Defence Secretary John Healey resigned in June, stating the government “had been unable to commit resources needed to defend the country at a time of rising threats.” The Bank of England has faced criticism over inflation forecasting accuracy. The UK Civil Service has been subject to repeated criticism over policy implementation.
These failures are distinct in origin and consequence. But they reflect a common pattern: institutional actors deploying new systems without adequate oversight; discovering failures through external pressure rather than internal competence; and failing to implement systemic reform in response.
What the CPS Must Do
The CPS must announce immediate institutional reforms:
- Mandatory verification protocols — All legal authorities cited in court documents must be verified against primary sources before filing, with AI-generated research flagged for mandatory secondary review.
- Audit of recent filings — A systematic review of all cases from the past 24 months involving AI-assisted legal research, with results published.
- Retraining requirements — Mandatory training for all counsel on AI risks, hallucination detection, and verification protocols.
- Public accountability — Disclosure of what disciplinary action, if any, was taken regarding the junior counsel’s conduct.
- Institutional reform — Structural changes to QA processes that would have caught this error before it reached court.
Without these steps, the message to the CPS and other prosecutorial agencies is clear: You can deploy powerful new technology, implement inadequate safeguards, submit false information to court, and face no meaningful consequences. That message serves no one except those who believe institutions should be held to lower standards than individuals.
Institutional Competence Requires Reform, Not Just Apology
The CPS’s admission that it fabricated legal authorities is not a technology story. It is an institutional failure story — one that reveals how a critical agency responsible for prosecuting serious crime has adopted tools without adequate oversight, failed basic quality control, and responded to exposure with apologies rather than reform.
The incident is instructive precisely because it was caught. A fabricated citation to the High Court is not a minor error. It is an abuse of the court’s process. That it was corrected does not negate the institutional failure that enabled it. The question now is whether the CPS will treat this as a signal that systemic safeguards must be implemented — or as an isolated incident to move past.
Institutional competence means learning from failure and implementing structural change. The CPS has a responsibility to demonstrate it remains capable of that standard.