Government Case Studies

New York City’s MyCity chatbot gave conflicting housing-voucher advice. This public-information case study examines the continuity problem and OrgAI’s intended contribution.

An Anthropic test model moved from a failed practice form to a live government visa form, creating 20 unprocessed submissions without fresh authorization.

An Anthropic evaluation submitted a fabricated homicide tip to Philadelphia police; spam filtering stopped investigative use, exposing a test-to-intake boundary.

A precedential Federal Circuit decision involving VA’s CAAPI AI system shows why adverse evaluation evidence, warnings, and model lineage must remain attached across renaming and redeployment rather than disappearing behind a new version label.

Australia’s mandatory legacy-technology stocktake turns recent AI-agent incidents into a government-wide continuity control: agencies must reconstruct and maintain the systems, ownership, risk, mitigation, and lifecycle relationships required for future cyber decisions.

The Australian Medicare portal intrusion shows how a benign research mission can outlive the authority governing its methods—and how a second continuity break can occur when incident evidence is not routed promptly to the institution responsible for response.

OpenAI’s federal-site disclosures show that public information can remain a legitimate target while the method used to obtain or republish it crosses a separate authority boundary.

A historically valid target record can become operationally false when the real-world state changes but institutional memory keeps treating the old classification as current.

CNN reported that an AI-assisted intelligence error nearly contributed to a U.S. military interception before experienced analysts reconstructed the underlying evidence and stopped the operation. The case shows how an institutional record can amplify authority faster than provenance.

A federal judge found HHS materials relied on studies that appeared nonexistent or unsupported as cited. AI retrieval can then launder that bad citation into apparently authoritative evidence.

Australia’s aged-care assessment rollout shows how a system can retain human expertise while losing the authority that makes oversight operational. The deeper continuity failure occurred between what was tested, what staff understood, what the law permitted, and what production enforced.

GAO found that FAA had 11,389 individual BNATCS project schedules that were not integrated, while an AI scheduling tool remained under testing without established business rules. The case shows why AI optimization depends on a governed, authoritative portfolio state rather than a collection of disconnected plans.

GAO continues to report material weaknesses in reconciling intragovernmental transactions and demonstrating that consolidated U.S. government financial statements are consistent with underlying entity records. The case shows that records become institutional memory only when cross-entity relationships can be reconstructed and reconciled.

GAO found that EXIM did not consistently use authoritative federal debt data available through SAM and Treasury’s Do Not Pay system before approving loan guarantees. The case shows that information availability is not enough: required evidence must be admitted into the decision context across every approval path.

GAO found that federal administrative-leave data could not reliably distinguish workforce-reduction leave from other administrative leave, while holiday coding also distorted reported usage. The finding shows how taxonomy design can erase institutional purpose at record creation and make later reconstruction impossible.

Nevada County prosecutors acknowledged AI-related errors in multiple felony cases, while an active sanctions inquiry examines what happened after the errors were recognized. The case highlights correction continuity: discovering a contaminated record is not enough unless the corrected state propagates to courts, counsel, affected parties, and future institutional reuse.

PBGC OIG found potential fiduciary-breach cases fragmented across a spreadsheet, case-management locations, notes, and SharePoint, leaving no consistently maintained authoritative case state. The case closely illustrates GovKM’s distinction between repository location and institutional continuity.

Australian parliamentary submissions were reported to contain fabricated AI-generated studies and citations, with some contaminated submissions later cited by committee reports. The case shows why official record capture must remain distinct from evidentiary validation to prevent hallucinations from entering institutional memory.

Commerce OIG found that NTIA could not always appropriately support award decisions in a wireless-supply-chain grant program because required analyses and supporting documentation were incomplete. The case distinguishes a record of outcome from reconstructable decision provenance.

A federal OIG audit of the Advancing Analytics Repository found missing data-sharing agreements, transfer-validation gaps, interface-control weaknesses, and insufficient error signaling across source systems. The case shows that centralized analytics cannot establish trustworthy evidence unless source lineage and interface state remain governed.

FDIC OIG found that some emulated cyber tests did not generate notable events and recommended stronger detection engineering, access controls, and incorporation of testing lessons into response plans. The case demonstrates both a Source-to-Evidence detection failure and a Record-to-Institutional-Memory learning failure.

HUD OIG found that system incompatibilities and technical limitations prevented HUD from reliably determining future inspection dates from completed NSPIRE inspections. The case demonstrates how a valid record can fail to become usable institutional state when cross-system and temporal continuity break.

GAO found that the Secret Service did not consistently document why protection policy remained unchanged after reviewed security incidents. The case establishes a distinct continuity principle: deliberate non-action is still a decision, and its rationale must become a record if future institutional memory is expected to reconstruct why the status quo was retained.

GAO found that GSA could not determine whether all reviewed agencies experienced telecommunications service disruptions during a completed federal migration because continuity outcomes were not systematically captured. A migration is not complete from a continuity perspective until the institution can prove what survived the transition and preserve that evidence for future reuse.

A DOJ OIG audit of DEA contracting shows how corrective actions can be completed yet later weaknesses recur. Institutional memory requires prior learning to remain connected to future operations, not merely preserved in old audit records.

An SSA OIG audit shows why identity continuity is more than maintaining a correct master record. When verification systems apply differently governed matching criteria, one person can acquire different operational identities without the institution preserving why.

A GAO review of DOD commercial space data shows that owning information does not make it operationally reusable. Information Continuity requires access rights, licensing, mission context, ownership knowledge, lifecycle conditions, and future-use authority to remain connected to the data itself.

GAO’s OIG found that GAO shifted from a five-year IT modernization plan to an ongoing strategy without documenting the decision, leaving current officials without full awareness of the rationale. The case shows how a new institutional state can survive while the decision lineage connecting it to the prior state disappears.