Johann Rehberger

Rehberger’s security leadership includes Electronic Arts, Microsoft Azure Data and Uber. At Microsoft he established and led the Azure Data red team as a principal security engineering manager. He also taught ethical hacking at the University of Washington. His writing connects those operational security experiences with adversarial testing of language-model agents.[4][6]

A safe-looking action can inherit an unsafe environment

His Claude Code case study examines a website-summary task that leads an agent into attacker-controlled files. The agent rejects a supplied executable but writes its own decoder; the surrounding directory changes how that otherwise ordinary code resolves dependencies. The lesson is that judging a generated command in isolation misses the environment and earlier steps that determine its effect. Rehberger reports small-sample targeted experiments, with successful defenses including static inspection and execution isolation.[3]

Build the team as well as the test

Cybersecurity Attacks: Red Team Strategies grew from his Azure Data experience. It covers team building, defensive tooling and collaboration between offensive and defensive teams, alongside technical foundations across operating systems. That combination treats a red team as a continuing organizational capability.[2]