Disclosed vulnerability
Sneaking invisible Unicode Tag instructions past developers in Windsurf
AIR-2025-0112 AIR record · 2025-08-windsurf-invisible-instructions
From the supporting source
But the key point is: invisible instructions can invoke tools.embracethered.com
The record
Disclosed
2025-08-23
Vendor
Windsurf / Codeium
Product
Windsurf Cascade (incl. SWE-1 model)
Risk domain
Agent behaviour › Unsafe actuation
Also touches
Data exfiltration channels
System type
Generative model
Who acted
Agent acted — Windsurf invoked tools from invisible Unicode instructions
Attack surface
Coding agent
Autonomy level
Semi-autonomous
Initial vector
Invisible unicode injection
Impact type
Data exfiltration, Unauthorized action
Confidence
High
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI01 – Agent Goal Hijack
- MITRE ATLAS
- AML.T0048 – External Harms
- NIST AI RMF
- MEASURE 2.6; MANAGE 1.3
- EU AI Act
- Annex III §1(c)
- ISO/IEC
- TR 24028; 24029-1; 23894
- AIUC-1
- C. Safety
- Scope
- agent misuse
- Mitigation
- Require confirmation for destructive actions; Implement dry-run modes; Use layered safety controls.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
Windsurf users processing untrusted files or tool output
Capability abused
Defense & remediation
Recorded defense
human review of file contents
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
unfixed at publication; disclosed 2025-05-30, Windsurf later said they would implement fixes
Taxonomy & references
Curator notes
Month of AI Bugs.
Supporting sources
Cite as
Enkrypt AI. "Sneaking invisible Unicode Tag instructions past developers in Windsurf". AIR, AIR-2025-0112. https://agentincident.vercel.app/i/AIR-2025-0112/ (accessed 2026-10-19).