Disclosed vulnerability
Microsoft fixes data exfiltration via image markdown injection in Azure AI Playground
AIR-2023-0022 AIR record · 2023-09-microsoft-azure-ai-playground-exfil
From the supporting source
When untrusted data makes it into the LLM prompt context it can instruct the model to inject an image markdown element.embracethered.com
The record
Disclosed
2023-09-29
Vendor
Microsoft
Product
Azure AI Playground
Risk domain
Privacy › Data exfiltration channels
Also touches
Confused deputy
System type
Generative model
Who acted
Agent acted — the playground rendered the exfiltration image
Attack surface
Enterprise assistant
Autonomy level
Assistant
Initial vector
Indirect prompt injection document
Impact type
Data exfiltration
Confidence
High
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI02 – Tool Misuse and Exploitation
- MITRE ATLAS
- AML.T0024 – Exfiltration via AI Inference API
- NIST AI RMF
- MAP 4.2
- EU AI Act
- Article 10
- ISO/IEC
- TR 24028; 23894
- AIUC-1
- A. Data & Privacy
- Scope
- tool misuse
- Mitigation
- Monitor network traffic for anomalous patterns; Restrict egress destinations.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
Azure AI Playground users
Capability abused
Defense & remediation
Recorded defense
none
Guardrail kinds
None
Guardrail outcome
Not applicable
Fix shipped
fixed 2023-09-13; reported 2023-07-24 via MSRC
Taxonomy & references
Curator notes
Johann Rehberger.
Supporting sources
Cite as
Enkrypt AI. "Microsoft fixes data exfiltration via image markdown injection in Azure AI Playground". AIR, AIR-2023-0022. https://agentincident.vercel.app/i/AIR-2023-0022/ (accessed 2026-10-19).