AML.CS0063
Prompt-Based Attacks Against Gemini via Calendar Invitations
- Type
- Exercise
- When
- 2025-08-06 to 2026-10-28
- Actor, as ATLAS names it
- Not stated
- Target
- Google Gemini
- ATLAS attributes it to SafeBreach Labs (security evaluation).
Techniques, in order
- 1Active ScanningReconnaissance · AML.T0006
- 2Discover AI Agent ConfigurationDiscovery · AML.T0084
- 3LLM Prompt CraftingAI Attack Adaptation · AML.T0065
- 4Prompt Infiltration via Public-Facing ApplicationInitial Access · AML.T0093
- 5LLM Prompt Injection: IndirectExecution · AML.T0051.001
- 6AI Agent Context Poisoning: ThreadPersistence · AML.T0080.001
- 7Delay Execution of LLM InstructionsDefense Evasion · AML.T0094
- 8LLM Prompt Injection: TriggeredExecution · AML.T0051.002
- 9LLM JailbreakDefense Evasion · AML.T0054
- 10External Harms: User HarmImpact · AML.T0048.003
- 11AI Agent Tool InvocationPrivilege Escalation · AML.T0053
- 12Data Destruction via AI Agent Tool InvocationImpact · AML.T0101
- 13Exfiltration via Cyber MeansExfiltration · AML.T0025
- 14Data from AI Services: AI Agent ToolsCollection · AML.T0085.001
- 15Exfiltration via AI Agent Tool InvocationExfiltration · AML.T0086
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.