Skip to content
The State of Exposed AgentsRead the report
Guard0

What is shadow AI?

Shadow AI is the use of AI tools, models and agents inside an organization without the knowledge or approval of the teams responsible for security and governance. It includes personal chatbot accounts used for work, coding agents and extensions installed on laptops, agents built inside business platforms, and agents deployed to cloud accounts without registration.

Also called shadow agents, unsanctioned AI. Last reviewed .

Why it matters

The early cases were about data leaving. In May 2023, TechCrunch reported that Samsung restricted generative AI tools on company devices after engineers uploaded internal source code to ChatGPT.

The newer cases are about software that acts. The OWASP MCP Top 10 lists MCP09:2025 Shadow MCP Servers, meaning MCP deployments nobody approved or supervises. NIST's AI Risk Management Framework expects the opposite: GOVERN 1.6 calls for mechanisms "to inventory AI systems".

Common risks

Data leaves in prompts and uploads
Whatever a person pastes into a personal chatbot account leaves the company, and nobody reviewed the terms that account runs under.
Nobody answers for it
An unapproved agent has no Human of Record: no one receives its findings, and no one retires it when the project ends.
It acts with real credentials
An agent holds tokens and grants and uses them without a person approving each call. What it can reach is a question of agent identity that nobody reviewed.
Unapproved MCP servers
The OWASP MCP Top 10 lists MCP09:2025 Shadow MCP Servers: MCP deployments nobody approved or supervises, each one new code with access to data.
It is missing from the inventory
A system that is not listed cannot be assessed. NIST's GOVERN 1.6 asks for mechanisms "to inventory AI systems".

Examples

Source code pasted into a chatbot
In May 2023, TechCrunch reported that Samsung restricted generative AI tools on company devices after engineers uploaded internal source code to ChatGPT.
An MCP server package nobody checked
On 25 September 2025, Postmark warned that an npm package named postmark-mcp, which Postmark did not publish, had added a hidden BCC in version 1.0.16 that copied every email it sent to an outside address. Anyone who had installed it was running an MCP server from an unknown author.
AI command-line tools on developer laptops
On 26 August 2025, malicious versions of several Nx packages were published to npm. According to the Nx postmortem, the post-install script searched the developer's machine for secrets and tried to use the AI command-line tools already installed there to help.
AI agent inventory
An AI agent inventory is what shadow AI is missing from.
Human of Record
A Human of Record is the first thing an unapproved agent lacks: a person who answers for it.
AI governance
AI governance sets the rule for what counts as approved.
MCP security
MCP security covers unapproved MCP servers, one of the fastest-growing kinds of shadow AI.

How Guard0 handles it

Guard0 finds AI tools and agents outside your inventory in three places: on laptops through your MDM, inside business platforms through their admin APIs, and in code and cloud accounts through read-only connections. Each one goes on the register with a person who answers for it and decides whether to approve it, replace it or retire it.

Shadow AI with Guard0

Read the guide: Shadow Agents: Finding the AI You Don't Know About

Questions

What is the difference between shadow AI and shadow IT?

Shadow IT is software nobody approved. Shadow AI is software nobody approved that also acts: it reads data, calls tools and holds credentials under the identity of a person. An inventory of applications answers shadow IT. Shadow AI needs what each agent can reach, what it did and who owns it. The shadow AI detection page sets the two side by side.

What counts as shadow AI?

Four things, all in use without approval: personal chatbot accounts used for work, coding agents and extensions installed on laptops, agents built inside business platforms, and agents deployed to cloud accounts without registration.

How is shadow AI found?

By reading the places AI runs instead of asking people what they use. Guard0 reads laptops through your MDM, business platforms through their admin APIs, and code and cloud accounts through read-only connections. See shadow AI detection.

Which frameworks ask for an AI inventory?

NIST's AI Risk Management Framework does, in GOVERN 1.6: "Mechanisms are in place to inventory AI systems and are resourced according to organizational risk priorities." An AI agent inventory is the list shadow AI is missing from.

Sources

  1. Samsung bans use of generative AI tools like ChatGPT after April internal data leak. TechCrunch.
  2. Information regarding malicious postmark-mcp package. Postmark.
  3. s1ngularity postmortem. Nx.
  4. OWASP MCP Top 10. OWASP.
  5. AI RMF Playbook: Govern. NIST.

Find the agents in your own company.

Start free: discover the AI on three people's machines, register one AI system and connect one repository or AWS account.