Chronicle YARA-L: AI Agent Indirect Context Injection & Markdown Exfiltration — Threat Hunting Query
Hunting Telemetry & Platform Metadata
Section titled “Hunting Telemetry & Platform Metadata”| Evaluation Dimension | Observed Telemetry / Attribute | Epistemic Status |
|---|---|---|
| Hunting Query Identifier | HUNT-CHRONICLE-LLM-PROMPT-INJECTION | Canonical Hunt Node |
| Target SIEM / Analytics Platform | Google Cloud Chronicle SecOps | Telemetry Ingestion Engine |
| Platform Syntax Dialect | chronicle_yaral | Query Language Standard |
| Telemetry & Log Sources | Application Gateway Telemetry, Agent Execution Traces, DNS Query Logs | Data Sensor Sources |
| Targeted Adversary Syndicates | CLOSEDQUORUM Threat Cluster | Threat Attribution |
| Associated Weaponized CVEs | CVE-2026-101861 | Vulnerabilities Targeted |
| Associated Intrusion Campaigns | CAMP-GLOVE-2026 | Monitored Campaign |
| Analytical Confidence | 95% | Hypothesis Validity Score |
| Primary Evidence Sources | SRC-CISCO-TALOS, SRC-GITHUB-ADVISORIES | Intelligence Attestation |
Threat Hunting Hypothesis
Section titled “Threat Hunting Hypothesis”Adversaries inject instructions into retrieved RAG documents that command autonomous agents to embed sensitive API keys inside markdown images fetched from attacker servers.
Adversary operations frequently slip through standard alerting thresholds by leveraging legitimate binaries or abusing valid service configurations. Proactive threat hunting formulates precise hypotheses based on known architectural chokepoints and queries historical telemetry to unearth stealthy intrusion footholds.
Production Hunting Query
Section titled “Production Hunting Query”rule hunt_llm_markdown_image_data_exfiltration { meta: description = "Hunts for outbound image rendering requests carrying base64 payloads initiated by AI agent runtime services" author = "Hermes Detection Engineering" severity = "HIGH" events: $http.target.url = /\!\s*\[.*?\]\(https?:\/\/[^\/]+\/[a-zA-Z0-9_-]+\.(?:png|jpg|gif)\?[a-zA-Z0-9_]+=([a-zA-Z0-9+\/=]{32,})\)/ $http.principal.process.command_line = /langflow|llama-index|langchain|autogen/ condition: $http}Triage Guidance & Investigation Protocol
Section titled “Triage Guidance & Investigation Protocol”Isolate the agent orchestration runner, review RAG ingestion history for recently indexed external web pages, and decode the base64 GET query parameters to identify stolen secrets.
- Verify Timeline Baseline: Review the first seen and last seen timestamps of returned events to distinguish chronic benign traffic from sudden operational spikes.
- Examine Parent-Child Lineage: Ensure that all parent processes match authorized deployment manifests and do not conceal unexpected interactive utilities.
- Inspect Network Pivots: Cross-reference communicating source and destination IP addresses against Hermes Adversary Infrastructure endpoints.