Disclosed vulnerability
Lack of isolation in agentic browsers resurfaces XSS- and CSRF-class harms
AIR-2026-0009 AIR record · 2026-01-trailofbits-agentic-browser-isolation
From the supporting source
Without proper isolation, these agents can be exploited to compromise any data or service the user's browser can reach.blog.trailofbits.com
The record
Disclosed
2026-01-13
Product
Unnamed agentic browsers (vendors declined coordinated disclosure)
Risk domain
Access control and permissions › Privilege escalation
Also touches
Data exfiltration channels, Confused deputy
System type
Generative model
Who acted
Agent acted — the agentic browsers reused authenticated sessions across origins
Attack surface
Browser agent
Autonomy level
Semi-autonomous
Initial vector
Indirect prompt injection web page
Impact type
Data exfiltration, Session confusion, Persistence
Confidence
High
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI03 – Identity and Privilege Abuse
- MITRE ATLAS
- AML.T0055 – Unsecured Credentials
- NIST AI RMF
- GOVERN 6.1
- EU AI Act
- Article 13
- ISO/IEC
- TR 24028; 42001; 23894
- AIUC-1
- B. Security
- Scope
- tool misuse
- Mitigation
- Apply least-privilege principle; Regular policy audits.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
Any data or service the user's browser can reach
Capability abused
Taxonomy & references
Curator notes
Author Lucas Bourtoule. Eight attack classes in three groups (manipulation, data exfiltration, persistence). Products unnamed because the affected vendors declined coordinated disclosure; no CVEs.
Supporting sources
Cite as
Enkrypt AI. "Lack of isolation in agentic browsers resurfaces XSS- and CSRF-class harms". AIR, AIR-2026-0009. https://agentincident.vercel.app/i/AIR-2026-0009/ (accessed 2026-10-19).