UNIFIED SOC INCIDENT RESPONSE BLUEPRINT
Autonomous AI Agents: How to Detect LiteLLM & Dify Exploits
Autonomous AI agents exploit CVE-2026-42271 and CVE-2026-48710 in LiteLLM and Dify to steal cloud API keys. Implement Starlette patches and hunt STS pivots.
Detection & Hunting Detectors (2)
[T1078.004] Adversary AI Agent Cloud Token Abuse via STSStandalone Rule
CloudAppEvents
| where Timestamp >= ago(7d)
| where ActionType in~ ("GetCallerIdentity", "AssumeRole", "CreateAccessKey")
| summarize CallCount = count() by AccountName, tostring(IPAddress), bin(Timestamp, 1h)
| where CallCount > 10[T1190] LiteLLM MCP Unauthenticated Remote Command ExecutionStandalone Rule
DeviceProcessEvents
| where Timestamp >= ago(7d)
| where InitiatingProcessFileName in~ ("python", "python3", "uvicorn")
| where InitiatingProcessCommandLine has_any ("litellm", "dify")
| where FileName in~ ("sh", "bash", "id", "whoami", "cat")
| project Timestamp, DeviceName, InitiatingProcessCommandLine, FileName, ProcessCommandLineVerified Indicators of Compromise (3)
Indicators of Compromise (IOCs)
3 Identified
TARGET SOFTWARE
LiteLLM, Dify, One API, Starlette frameworkURL
POST /mcp-rest/test/tools/listUSER-AGENT
Python/3.14 aiohttp/3.14.5