Home / Glossary / Shadow agents
Glossary
What are shadow agents?
Shadow agents are AI agents that employees run at work without security's approval. They are the part of shadow AI that can act: they run commands, read files and call systems with the employee's access. Many run on the device and never touch the network, so they do not show up in a gateway log.
By the AIBound team · Updated 17 November 2026
Key takeaways
- Shadow agents are AI agents that employees run at work without security's approval.
- A shadow agent can act on its own, with the access of the employee who runs it.
- Many shadow agents run on the device, not on the network, so they never show up in a gateway log.
Why shadow agents matter
An unapproved chatbot can take in data that an employee pastes into it. An unapproved agent can go and get the data itself. It can read files, run commands and use the employee's credentials, and nobody reviewed what it may touch. Each MCP server or skill the employee adds gives it more reach.
Shadow agents also last. A personal agent can keep running between sessions and long after the task that started it. No one owns it, updates it or removes it. When an auditor asks which agents can reach company systems, there is no list to show.
How shadow agents appear
A shadow agent is rarely a bad act. It starts when an employee wants to get work done faster.
- 1 An employee installs an AI agent on a work device, or sets up a personal agent with a personal account.
- 2 The employee connects it to files, code, email or company systems, often through an MCP server or a skill.
- 3 The agent acts with the employee's access, and security sees nothing because it runs on the device.
- 4 The agent keeps running, and its reach grows each time something new is connected.
Shadow agents vs shadow AI
Shadow AI is any AI that employees use without approval. Shadow agents are the part of it that takes actions.
| Question | Shadow agents | Shadow AI |
|---|---|---|
| What is it? | Unapproved AI agents | Any unapproved AI: agents, models, skills and apps |
| What can it do? | Act on its own and reach files and systems | Ranges from answering a prompt to acting on its own |
| What is the worst case? | An agent acts on company systems with nobody watching | Company data leaves through a tool nobody reviewed |
Sources
- Cloud Security Alliance research note: The AI Agent Governance Gap · labs.cloudsecurityalliance.org
- OWASP LLM06: Excessive Agency · genai.owasp.org
Sources checked 19 November 2026
Examples
Shadow agent examples
FAQ
Frequently asked questions
What are shadow agents?
Shadow agents are AI agents that employees run at work without security's approval. They include coding agents, desktop assistants and personal agents. Unlike a chatbot, an agent takes actions on its own, with the access of the employee who runs it, so nobody has reviewed what it can reach.
What is the difference between shadow agents and shadow AI?
Shadow AI is the wider term. It covers every AI that employees use without approval: agents, MCP servers, skills, local models and personal-account apps. Shadow agents are the part that acts. A chat app waits for a prompt, and an agent runs commands and reads files by itself.
Why are shadow agents a risk?
A shadow agent works with the employee's access, and nobody checked its limits. It can read credentials, change files or send data to an outside host. Each MCP server or skill added to it widens its reach. Because security does not know it exists, there is no rule on it and no record for an audit.
How do you find shadow agents?
Look at the device, because many agents never touch the network. AIBound installs on employee devices in 15 minutes and finds the agents, MCP servers, skills and models in real time, with the person who runs each one. You then choose what to allow, ask about or deny.