Disclosed vulnerability
CVE-2026-79696: unauthenticated RCE in Google ADK for Python (adk web) via crafted test session replay
AIR-2026-0230 AIR record · 2026-09-google-adk-python-test-replay-rce
From the supporting source
allows an unauthenticated remote attacker to execute arbitrary code using a crafted test session replay.cveawg.mitre.org
The record
Disclosed
2026-09-09
Vendor
Google
Product
Google Cloud Agent Development Kit (ADK) for Python
Risk domain
Access control and permissions › Privilege escalation
System type
Generative model
Who acted
AI was the target — agent framework RCE; no model decision
Attack surface
Agent framework
Autonomy level
Assistant
Initial vector
Unauthenticated network request
Impact type
Remote code execution
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
Google ADK for Python 2.0.0 through 2.6.0 (OSS, Cloud Run, GKE) where pytest is installed
Capability abused
Defense & remediation
Recorded defense
standard-library denylist
Guardrail kinds
Permission scope
Guardrail outcome
Silent
Fix shipped
google-adk 2.7.0
Taxonomy & references
Curator notes
CVSS 10.0 (v4.0). Published 2026-09-09; reported by Sanil Dulal. Fixed in ADK 2.7.0 (commit a16f6da). PoC/disclosure; no real victim. Distinct from CVE-2026-79707 (ADK builder path traversal, 8.7).
Supporting sources
Cite as
Enkrypt AI. "CVE-2026-79696: unauthenticated RCE in Google ADK for Python (adk web) via crafted test session replay". AIR, AIR-2026-0230. https://agentincident.vercel.app/i/AIR-2026-0230/ (accessed 2026-10-19).