What is shadow AI?
Shadow AI is the use of AI tools, models, or features at work without the knowledge or approval of IT, security, or governance owners. It is the AI form of shadow IT, and it spreads faster because most tools are free and work in a browser. Staff usually adopt them because the work gets faster and no approved option exists.
Shadow AI takes several common forms:
- Employees paste contracts, customer records, or source code into personal accounts on public chatbots.
- Meeting note-taker bots and browser extensions capture calls and documents and send them to their vendors.
- Vendors switch on AI features inside approved software in a routine update.
- Staff approve OAuth requests that give AI apps lasting access to their work mailbox and files.
- Teams build scripts and no-code automations that call model APIs with personal keys.
Why shadow AI matters
The main risk is data leaving the company’s control. Depending on the account terms, content entered into a consumer tool may be retained, read by the vendor’s staff, or used for training. That can breach client confidentiality terms, data processing agreements, and laws such as HIPAA and the GDPR. In security terms, this is unmanaged data exfiltration. Work product also lands in personal accounts that leave with the employee and fall outside legal holds. A company cannot govern systems it does not know exist, so finding shadow AI is usually the first step in building an AI inventory and an AI acceptable use policy.
Shadow AI in practice
A 200-person law firm reviews the OAuth grants in its identity provider and finds that dozens of lawyers connected an AI note-taker to their calendars. The bot has been joining client calls and storing transcripts on its vendor’s servers. The firm revokes the grants, blocks the note-taker, and offers an approved transcription tool that runs on infrastructure the firm controls, respects ethical walls, and logs each request by user and matter.
In Kirk Tech Solutions engagements, the approved alternative is usually a private assistant on open-weight models with single sign-on and per-user logging, built to be as convenient as the personal accounts it replaces.
What to watch for
- Blocking public tools without an approved alternative pushes use onto personal phones, where nobody can see it.
- Review OAuth grants and browser extensions regularly, because connected AI apps keep their access after people stop using them.
- Data loss prevention rules and an AI gateway can catch sensitive data bound for known AI services, but only on company networks and devices.
- Treat regulated data found in an unapproved tool as a possible incident under the applicable law, in addition to a policy violation.