Shadow AI

Employees using AI tools or AI agents for work that their organization has not approved or does not know about.

For example, a staff member might paste customer records into a public chatbot to summarize them, or connect an unapproved AI agent to a work inbox or customer database to automate a task.

The use of AI applications, models, browser extensions, or agents outside an organization's security and IT governance, which means they run without security review, access controls, logging, or contractual data protections. Shadow AI is a subset of shadow IT, the broader term for any technology used at work without approval.

Shadow AI usually shows up in a few forms:

• Public generative AI tools, where staff paste or upload company or customer data into a consumer chatbot.
• AI features inside approved software, switched on without review, that send data to a new third party.
• Unapproved agents and connectors, which staff link to email, files, or customer systems using their own login so the agent can act for them.

Organizations usually manage it in four steps:

1. Discover which AI tools are in use, through network, browser, and login records.
2. Classify each tool by the data it touches and the access it holds.
3. Approve, restrict, or block each one, and offer approved alternatives.
4. Monitor for new tools and for agents acting with employee credentials.

Shadow AI sits next to terms that are easy to mix up:

• Data breach is one outcome shadow AI can cause, when customer data reaches a tool that exposes or keeps it.
• Authorization decides what an approved party may do, and a shadow agent typically borrows an employee's permissions with no separate check.