An approved agent with an unapproved skill isnโt fully governed.
Shadow AI brought unapproved AI into the workplace. Shadow Skills is its agentic evolution: ungoverned instructions that guide what approved agents do.
Skills are reusable blueprints that tell an AI agent how to do its job. They influence what data an agent accesses and which actions it takes, using permissions it already has. A skill may be approved initially, but it can be maliciously compromised later. This can leave a trusted agent following instructions that lead to data leaks or harmful code execution.
Security controls may show an agent taking an allowed action without revealing the skill that directed it. That makes it difficult for security teams to judge whether an apparently legitimate action is following instructions the organization can trust.
This book explains the risks agent skills introduce and provides a framework for governing their use.
Whatโs inside: