AML.CS0052
LLMSmith: RCE Vulnerabilities in LLM-Integrated Applications
- Type
- Exercise
- When
- 2025-02-27
- Actor, as ATLAS names it
- Researchers at University of Chinese Academy of Sciences, Shandong University, and University of New South Wales
- Target
- LLM Integration Frameworks
Techniques, in order
- 1Develop CapabilitiesResource Development · AML.T0017
- 2Search Application RepositoriesReconnaissance · AML.T0004
- 3Discover AI Agent Configuration: Call ChainsDiscovery · AML.T0084.003
- 4LLM Prompt CraftingAI Attack Adaptation · AML.T0065
- 5Exploit Public-Facing ApplicationInitial Access · AML.T0049
- 6LLM Prompt Injection: DirectExecution · AML.T0051.000
- 7LLM JailbreakDefense Evasion · AML.T0054
- 8AI Agent Tool InvocationPrivilege Escalation · AML.T0053
- 9Command and Scripting InterpreterExecution · AML.T0050
- 10Escape to HostPrivilege Escalation · AML.T0105
- 11Cyber Communication ChannelCommand and Control · AML.T0072
- 12Machine Compromise: Local AI AgentImpact · AML.T0112.000
From MITRE ATLAS 2026.10 (Apache-2.0). ShieldMap lists who and what was involved and the techniques used, and links to ATLAS for the write-up; it does not copy the steps. Attribution is ATLAS's, not ShieldPi's.