Microsoft Research: AI Can Generate Realistic Command-Line and Process Telemetry

Microsoft researchers demonstrate an AI-driven approach to synthesize realistic command-line and process telemetry from attacker tactics (e.g., MITRE ATT&CK) to accelerate detection engineering. The work evaluates prompt-based generation, multi-agent (agentic) workflows, and reinforcement learning with verifiable rewards (RLVR), finding agentic workflows most effective at producing high-fidelity logs; synthetic telemetry promises safer, reproducible, and scalable datasets for testing and improving detection systems without executing real attacks.