Disclosed vulnerability
HuggingChat: zero-click data exfiltration via unvalidated markdown image rendering; disclosed publicly after HuggingFace did not respond
AIR-2025-0183 AIR record · 2025-12-huggingchat-markdown-image-exfil
From the supporting source
This vulnerability was responsibly disclosed to HuggingFace on 12/1/2025. As HuggingFace's security team has not responded to the disclosure, we are disclosing publicly to ensure users are informed of the risks and can take precautions accordingly.promptarmor.com
The record
Disclosed
2025-12-01
Vendor
Hugging Face
Product
HuggingChat
Risk domain
Privacy › Data exfiltration channels
Also touches
Confused deputy
System type
Generative model
Who acted
Agent acted — HuggingChat rendered the exfiltration image
Attack surface
Consumer assistant
Autonomy level
Assistant
Initial vector
Indirect prompt injection web content
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
HuggingChat users who add untrusted documents, webpages or MCP tools to context
Capability abused
Defense & remediation
Recorded defense
none
Guardrail kinds
None
Guardrail outcome
Not applicable
Fix shipped
unpatched at disclosure; vendor did not respond
Taxonomy & references
Curator notes
PromptArmor. date_disclosed here is the date of the private report to HuggingFace (2025-12-01); the public disclosure date is not stated on the page. Agentic hook: MCP tool responses are named as one of the injection channels.
Supporting sources
Cite as
Enkrypt AI. "HuggingChat: zero-click data exfiltration via unvalidated markdown image rendering; disclosed publicly after HuggingFace did not respond". AIR, AIR-2025-0183. https://agentincident.vercel.app/i/AIR-2025-0183/ (accessed 2026-10-19).