Disclosed vulnerability
OpenAI's own API log viewer renders model-generated markdown images, exfiltrating data when a developer reviews a flagged conversation
AIR-2026-0016 AIR record · 2026-01-openai-api-logs-markdown-exfil
From the supporting source
OpenAI's API log viewer is vulnerable to a data exfiltration attack, exposing apps and agents that use OpenAI APIs, even if developers (and Agent Builder users) leverage all available defenses.promptarmor.com
The record
Disclosed
2026-01-22
Vendor
OpenAI
Product
OpenAI Platform API logs (Responses/Conversations APIs, Agent Builder, Assistants)
Risk domain
Privacy › Data exfiltration channels
Also touches
Confused deputy
System type
Generative model
Who acted
Agent acted — the vendor console rendered model-generated markdown images, fetching them
Attack surface
Agent framework
Autonomy level
Assistant
Initial vector
Indirect prompt injection data source
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
any application or agent built on OpenAI APIs whose logs a developer reviews
Capability abused
Defense & remediation
Recorded defense
application-level output blocking by the developer
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
report reopened 2026-01-22, fix underway
Taxonomy & references
Curator notes
PromptArmor. Notable because the sink is the vendor's own console, not the application: an app that correctly blocks the malicious output still leaks when a human reviews the flagged conversation in OpenAI's log viewer. Initial disclosure 2025-11-17, closed 'Not Applicable' 2025-12-04, reopened 2026-01-22.
Cite as
Enkrypt AI. "OpenAI's own API log viewer renders model-generated markdown images, exfiltrating data when a developer reviews a flagged conversation". AIR, AIR-2026-0016. https://agentincident.vercel.app/i/AIR-2026-0016/ (accessed 2026-10-19).