What is shadow AI?
Shadow AI is the AI-specific form of shadow IT: staff using AI services that the organization never approved, reviewed or even knows about. An analyst pastes a customer spreadsheet into a free chatbot to get a summary, a developer connects an unvetted coding assistant to company repositories, a manager installs a browser extension that rewrites email, or a meeting assistant joins calls and stores the transcripts in its vendor's cloud. The motive is almost always speed rather than malice, which is why shadow AI is common and easy to miss.
Why shadow AI is a risk
- Data leaves the organization. Whatever is pasted or uploaded now sits with a third party, outside retention rules, legal holds and breach monitoring. Some consumer AI services keep conversations and may use them to improve their models unless the user opts out.
- No vendor review. Nobody checked where the provider stores data, who can read it, or what happens if it is breached.
- Compliance exposure. Personal data sent to an unapproved AI service can breach privacy law, customer contracts or data residency commitments.
- Broad access grants. AI extensions and plugins often ask to read every page or a whole mailbox through OAuth, and keep that access long after the user forgets them.
- Unreviewed output. Code, contract language or figures produced by an unapproved tool reach production without any check.
The pattern drew wide attention in 2023, when Samsung restricted generative AI tools after employees pasted source code and internal meeting notes into a public chatbot.
How organizations find shadow AI
Discovery uses the same sources as shadow IT. A cloud access security broker or secure web gateway lists the AI services reached from company devices, and a DLP alert might read action=block category=generative-ai dest=chat.example-ai.com content=source_code user=jlee. Identity provider logs show which AI apps users have granted OAuth access to mail or files, and expense reports reveal paid subscriptions nobody approved.
Managing it without driving it underground
Blocking alone tends to push use onto personal phones, where security sees nothing at all. A workable program combines:
- an AI acceptable use policy that names approved tools and the data classifications allowed in each
- a sanctioned enterprise AI service whose contract excludes training on company data and provides SSO, audit logs and retention controls
- DLP on endpoints and web traffic that warns or blocks when source code, card numbers or labeled documents head to AI sites
- administrator approval for AI browser extensions and OAuth app grants
- a quick way to request a new tool, plus training on why the rules exist
Shadow AI vs. shadow IT
All shadow AI is shadow IT, but it adds two problems. Data given to an AI tool may be retained or used for training, so it can resurface later, and AI tools produce work and sometimes take actions that people rely on without review. The controls come from the same family: discovery, a sanctioned alternative, DLP and policy.
Shadow AI on the CompTIA Security+ exam
SY0-801 places shadow IT among the vulnerability types in objective 2.4, and objective 2.6 covers the data loss and privacy risks of using AI. Shadow AI sits where the two meet. A scenario in which employees paste confidential data into a public chatbot, or a team adopts an AI tool that IT never reviewed, points here. The best answers pair an AI acceptable use policy and DLP enforcement with an approved enterprise tool; banning all AI with no alternative is usually the weaker choice. Distractors include a malicious insider, which implies intent to cause harm, and prompt injection, which requires an attacker.