Safety failure
Realized harm
Microsoft Copilot told users that court reporter Martin Bernklau was a child abuser and fraudster, conflating him with defendants he had covered, and published his address
AIR-2024-0034 AIR record · 2024-08-microsoft-copilot-bernklau-defamation
A real party was affected. This incident caused actual consequences, not a demonstrated capability.
From the supporting source
Copilot falsely claimed Bernklau had been charged with and convicted of child abuse and exploiting dependents.theregister.com
The record
Disclosed
2024-08-16
Occurred
2024-08
Vendor
Microsoft
Product
Microsoft Copilot
Risk domain
Privacy › Sensitive data exposure
Also touches
Human manipulation
System type
Generative model
Who acted
Agent acted (contested) — the assistant published defamatory statements and a home address to real users
Attack surface
Consumer assistant
Autonomy level
Assistant
Initial vector
None autonomous action
Impact type
Reputational harm, Privacy violation
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
one named journalist; the failure mode reaches every reporter, lawyer and judge whose name co-occurs with the crimes they document
Capability abused
Defense & remediation
Recorded defense
manual suppression of the specific entries
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
partial; Microsoft's removals were temporary and the output reappeared
Taxonomy & references
Curator notes
Included despite being text output because it turns on retrieval and grounding in a deployed assistant, and because the remediation failure is the point: Microsoft suppressed the specific claims and the model regenerated them. There is no delete for an inference. Bernklau's contact details were emitted alongside the accusations. AI Incident Database incident 770.
Supporting sources
Cite as
Enkrypt AI. "Microsoft Copilot told users that court reporter Martin Bernklau was a child abuser and fraudster, conflating him with defendants he had covered, and published his address". AIR, AIR-2024-0034. https://agentincident.vercel.app/i/AIR-2024-0034/ (accessed 2026-10-19).