Research demo
Escape from Pod 9: poisoned telemetry makes an AI SRE agent deploy a privileged ransomware container on EKS
AIR-2026-0203 AIR record · 2026-08-straiker-pod9-sre-agent-ransomware
From the supporting source
No software vulnerability was exploited. No credentials were stolen. No zero-day was used.straiker.ai
The record
Disclosed
2026-08-24
Vendor
Amazon Web Services (Bedrock) / custom agent
Product
AI SRE agent built on Amazon Bedrock monitoring an EKS workload via OpenTelemetry
Risk domain
Access control and permissions › Privilege escalation
Also touches
Confused deputy
System type
Generative model
Who acted
Agent acted — the SRE agent deployed the privileged ransomware container
Attack surface
Autonomous operations
Autonomy level
Semi-autonomous
Initial vector
Indirect prompt injection telemetry
Impact type
Ransomware deployment, Privilege escalation, Cluster compromise
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
Kubernetes clusters where an AI SRE agent holds deployment permissions and ingests untrusted telemetry
Capability abused
Taxonomy & references
Curator notes
Author Eduardo Arrucha (Straiker STAR Labs). Injection carried in failed-login records flowing through OpenTelemetry into the agent's context. Straiker is explicit that the flaw is agent design, not Bedrock or the model.
Cite as
Enkrypt AI. "Escape from Pod 9: poisoned telemetry makes an AI SRE agent deploy a privileged ransomware container on EKS". AIR, AIR-2026-0203. https://agentincident.vercel.app/i/AIR-2026-0203/ (accessed 2026-10-19).