news briefAI in CourtsAI Governance

US Federal Circuit revives whistleblower case over VA AI claims tool

The Federal Circuit reversed an MSPB jurisdiction dismissal over disclosures about a VA AI claims tool, without deciding retaliation merits.

Edited and verified by SafeLegalAI

Researched with AI assistance · sources verified by SafeLegalAI · How this was made ↓

The US Court of Appeals for the Federal Circuit revived a whistleblower appeal on 22 September 2026 over disclosures about a Department of Veterans Affairs artificial-intelligence claims tool.

In a precedential US federal decision, the court held that Dr Babatunde Oguntade had made a non-frivolous allegation that his disclosures about the VA’s Claims Attribute Application Programming Interface, or CAAPI, evidenced gross mismanagement. The court reversed the Merit Systems Protection Board’s jurisdiction dismissal and remanded; it did not decide whether retaliation occurred.

What did the court decide?

The opinion says Dr Oguntade was a Presidential Innovation Fellow assigned in 2020 to update CAAPI, an AI tool intended to reduce the time veterans waited for disability-benefits decisions. The VA had associated the tool with a three-to-five-day reduction in wait time and $1.5 million in direct labour-cost savings.

After reviewing claims data, Dr Oguntade told VA officials that CAAPI was “giving wrong predictions” and that the agency’s claims about the tool were “not tethered to the truth”. The court recorded his allegation that CAAPI slowed processing by five days on average and by nine weeks for complicated submissions.

The Federal Circuit held that those allegations were enough for jurisdiction under the whistleblower statute. The panel wrote that Dr Oguntade had plausibly alleged serious agency error when the VA continued using CAAPI on live claims after his warnings, without changing the error-prone classifier.

What did the court leave open?

The ruling is a jurisdiction decision, not a final finding that CAAPI was defective or that the VA retaliated unlawfully. The panel said it was not deciding whether Dr Oguntade would ultimately win his Whistleblower Protection Act claim.

Judge Lourie’s concurrence added a caution about over-reading the opinion. He wrote that the case should not discourage experimentation aimed at the VA claims backlog, while agreeing that disclosures about problems in an experimental programme can, on these facts, support a gross-mismanagement theory at the pleading stage.

What follows from this decision?

  • The case returns to the MSPB for further proceedings; no damages or merits remedy was ordered by the Federal Circuit.
  • The opinion treats AI-performance claims in a benefits-adjudication system as potentially material to whistleblower jurisdiction when the alleged errors affect live claims.
  • The court described CAAPI as an AI tool, but the opinion does not call it generative AI and does not involve fabricated legal citations.

The decision belongs with SafeLegalAI’s public-sector AI coverage, rather than the fabricated-citation tracker, unless a separate record is later created for CAAPI.

Sources