The Normalization of Deviance in AI: Lessons From and Beyond the Month of AI Bugs

In August 2025, Johann published daily research on Embrace The Red as part of the Month of AI Bugs, uncovering critical vulnerabilities in agentic AI systems, particularly computer-use and coding agents. This talk shares the testing methodology behind that research, highlights severe findings like RCE, credential exfiltration, persistent prompt injection, and “ZombAIs,” and argues that vendors’ repeated reintroduction of these flaws signals a growing normalization of deviance that shifts risk onto end users.