Disclosed vulnerability
Hijacking Windsurf: prompt injection leaks developer secrets via read_url_content and image rendering
AIR-2025-0110 AIR record · 2025-08-windsurf-data-exfiltration
From the supporting source
As demonstrated, weaknesses in tool invocation and image rendering allow a third-party attacker to embed malicious instructions into source code, websites, or even through RAG poisoning.embracethered.com
The record
Disclosed
2025-08-21
Vendor
Windsurf / Codeium
Product
Windsurf Cascade
Risk domain
Privacy › Data exfiltration channels
Also touches
Credential theft, Confused deputy
System type
Generative model
Who acted
Agent acted — Windsurf fetched URLs and rendered images to leak secrets
Attack surface
Coding agent
Autonomy level
Semi-autonomous
Initial vector
Indirect prompt injection source code
Impact type
Data exfiltration, Credential theft
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
Windsurf developer workstations
Capability abused
Defense & remediation
Recorded defense
none - read_url_content makes outbound requests without approval
Guardrail kinds
None
Guardrail outcome
Not applicable
Fix shipped
unfixed at publication; Windsurf said they would work on fixes, no ETA
Taxonomy & references
Curator notes
Month of AI Bugs. Disclosed 2025-05-30.
Supporting sources
Cite as
Enkrypt AI. "Hijacking Windsurf: prompt injection leaks developer secrets via read_url_content and image rendering". AIR, AIR-2025-0110. https://agentincident.vercel.app/i/AIR-2025-0110/ (accessed 2026-10-19).