A Guth Labs publication ↗
Back to News

Evoke launches lab for investigating AI agent incidents

Green glass shield crossed by a gold trail connecting cream and sage evidence nodes.

Evoke Security launched Defender CTF on November 17, a hands-on training lab for investigating incidents involving AI agents. The company announcement describes a staged incident modeled on behavior it has observed in customer environments.

Participants receive records of prompts, instructions, tool and Model Context Protocol calls, endpoint commands, accessed files and actions across cloud services and business applications. The exercise connects those records to the responsible person and agent, then asks investigators to reconstruct a timeline.

The central task is separating the user’s request from the agent’s actual behavior and identifying where it exceeded its instructions. Evoke says participants also see which parts of the investigation disappear when agent-specific telemetry is missing.

The lab targets incident responders, security operations analysts, threat hunters and detection engineers. The announcement does not provide independent evidence of training effectiveness or establish the company’s claim that this is the first such exercise.

Its practical contribution is a concrete investigation workflow: examining agent instructions and intermediate tool activity alongside conventional endpoint and identity records, rather than assuming those older logs explain the entire incident.

Sources and citations

Revision history

Published .