AML.CS0066
ZombieAgent: Data Exfiltration Attack on ChatGPT
- Type
- Exercise
- When
- 2025-09-25
- Actor, as ATLAS names it
- Not stated
- Target
- OpenAI ChatGPT
- ATLAS attributes it to Radware (security evaluation).
- On ShieldMap: ChatGPT (matched by name in the target).
Techniques, in order
- 1LLM Prompt CraftingAI Attack Adaptation · AML.T0065
- 2Stage CapabilitiesResource Development · AML.T0079
- 3LLM Prompt ObfuscationDefense Evasion · AML.T0068
- 4Prompt Infiltration via Public-Facing ApplicationInitial Access · AML.T0093
- 5LLM Prompt Injection: IndirectExecution · AML.T0051.001
- 6AI Agent Context Poisoning: MemoryPersistence · AML.T0080.000
- 7AI Agent Tool InvocationPrivilege Escalation · AML.T0053
- 8Data from AI Services: AI Agent ToolsCollection · AML.T0085.001
- 9LLM JailbreakDefense Evasion · AML.T0054
- 10Exfiltration via AI Agent Tool InvocationExfiltration · AML.T0086
- 11LLM Prompt PropagationLateral Movement · AML.T0143
How these attacks work
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.