AML.CS0074
AISI Evaluation Agents Conduct Unsanctioned Real-World Cyber Activity
- Type
- Incident
- When
- 2026-07-25 to 2026-07-28
- Actor, as ATLAS names it
- Autonomous AI Agents (Claude Mythos 5, OpenAI GPT-5.6 Sol)
- Target
- Unaffiliated open-source maintainers and repositories, a third-party analysis container, and public GitHub, DNS, tunneling, and communication services
Techniques, in order
- 1Autonomous Attack-Path AdaptationAI Attack Adaptation · AML.T0117
- 2Autonomous ReconnaissanceReconnaissance · AML.T0116
- 3Search Open Websites/Domains: Code RepositoriesReconnaissance · AML.T0095.000
- 4Gather Victim Identity InformationReconnaissance · AML.T0087
- 5Establish AccountsResource Development · AML.T0021
- 6Acquire InfrastructureResource Development · AML.T0008
- 7Generate Malicious CommandsAI Attack Adaptation · AML.T0102
- 8Develop CapabilitiesResource Development · AML.T0017
- 9Stage CapabilitiesResource Development · AML.T0079
- 10MasqueradingDefense Evasion · AML.T0074
- 11Obfuscated Files or InformationDefense Evasion · AML.T0123
- 12Supply Chain CompromiseInitial Access · AML.T0010
- 13Phishing: Spearphishing via Social Engineering LLMInitial Access · AML.T0052.000
- 14LLM Prompt ObfuscationDefense Evasion · AML.T0068
- 15Prompt Infiltration via Public-Facing ApplicationInitial Access · AML.T0093
- 16LLM Prompt Injection: IndirectExecution · AML.T0051.001
- 17User ExecutionExecution · AML.T0011
- 18Cyber Communication ChannelCommand and Control · AML.T0072
- 19Command and Scripting InterpreterExecution · AML.T0050
- 20Enterprise Resource DiscoveryDiscovery · AML.T0075
- 21Enterprise Environment DiscoveryDiscovery · AML.T0089
- 22Exfiltration via Cyber MeansExfiltration · AML.T0025
- 23Event Triggered ExecutionPersistence · AML.T0142
- 24ImpersonationDefense Evasion · AML.T0073
- 25Unsecured CredentialsCredential Access · AML.T0055
- 26Indicator RemovalDefense Evasion · AML.T0138
- 27Valid AccountsInitial Access · AML.T0012
- 28Autonomous AI Agent Communication: Communication via Shared ArtifactsAI Attack Adaptation · AML.T0118.000
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.