Agent Incident RegistryAIR
v1 · updated 19 Oct 2026
In the wild Realized harm

NSW Reconstruction Authority contractor uploaded flood-victim personal and health data for ~3,000 people to ChatGPT

AIR-2025-0091 AIR record · 2025-03-nsw-reconstruction-authority-chatgpt-upload
A real party was affected. This incident caused actual consequences, not a demonstrated capability.
From the supporting source
The breach occurred when a former contractor of the RA uploaded data containing personal information to an unsecured AI tool which was not authorised by the department.news.com.au
The record
Disclosed
2025-08-11
Occurred
2025-03
Vendor
OpenAI / NSW Reconstruction Authority (deployer)
Product
ChatGPT
Risk domain
Privacy › Sensitive data exposure
System type
Generative model
Who acted
AI was the target (contested) — the AI service was the passive recipient in a contractor-caused input disclosure; no model output or autonomous action caused the loss
Attack surface
Enterprise assistant
Autonomy level
Assistant
Initial vector
None autonomous action
Impact type
Privacy violation, Information disclosure
Confidence
High
Submitted by
EAI Internal
Crosswalk
OWASP Agentic Top 10 (2026)
ASI06 – Memory & Context Poisoning
MITRE ATLAS
AML.T0057 – LLM Data Leakage
NIST AI RMF
MEASURE 2.10
EU AI Act
Article 10
ISO/IEC
TR 24028; 23894
AIUC-1
A. Data & Privacy
Scope
tool misuse
Mitigation
Use data anonymization; Implement log sanitization to scrub PII.
Taxonomy
Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
up to 3,000 Northern Rivers Resilient Homes Program applicants; email addresses, phone numbers, personal and health information
Capability abused
file_uploadretrievaldata_retention
Defense & remediation
Recorded defense
departmental policy prohibiting unauthorised AI tools
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
Cyber Security NSW investigation and dark-web monitoring; affected individuals notified
Taxonomy & references
OWASP ASI02 Tool Misuse and Exploitationinsider_unintentionalshadow_aiunauthorised_toolgovernment_datahealth_data
Curator notes
Government-sector counterpart to the Samsung case, with a sharper victim class: disaster-recovery applicants whose health data was already sensitive. Primary source is the agency's own breach notice. Exfiltration is unconfirmed but not excluded, which is exactly the position an organisation is left in once data leaves for a third-party model. AI Incident Database incident 1228.
Cite as
Enkrypt AI. "NSW Reconstruction Authority contractor uploaded flood-victim personal and health data for ~3,000 people to ChatGPT". AIR, AIR-2025-0091. https://agentincident.vercel.app/i/AIR-2025-0091/ (accessed 2026-10-19).