Learn More About Us
Shadow AI is the use of AI tools by employees without approval from IT or security. It is everywhere, and it is expensive. IBM’s 2025 research found that breaches involving heavy shadow AI cost $670,000 more than average. The threat is rarely a bad actor. It is a helpful employee pasting sensitive data into a public chatbot to save a few minutes.
Shadow AI is any AI tool, model, or app used for work without IT or security approval. It ranges from one person pasting source code into a free chatbot to a whole team running an unvetted AI plugin on customer data. The driver is almost never malicious. It is the pull of getting work done faster.
Free, browser-based tools made this easy. An employee who finds the approved software slow or limited can open ChatGPT or Gemini in seconds, with no procurement process to wait on. The numbers back this up. 43% of employees admit sharing sensitive work data with AI tools without employer knowledge (NCA/CybSafe 2025).
Shadow AI is not the same as shadow IT. Shadow IT is about unsanctioned software or hardware. Shadow AI adds a sharper problem: external models can process, learn from, and retain whatever data an employee feeds them. Once proprietary information goes into a public model, you lose control over where it lives and how it gets reused.
The biggest risk is data leakage. When customer records, source code, or strategy documents go into a public model, that data leaves your control. In 2023, Samsung engineers leaked proprietary semiconductor source code and internal meeting notes into ChatGPT. The company banned the tool and started building its own.
Regulated businesses carry a second layer of risk. Feeding patient records, client files, or payment data into an unvetted tool can breach GDPR, HIPAA, or CCPA. Shadow AI breaches hit the most sensitive data hardest: AI-related security incidents involving shadow AI are more likely to expose PII (65%) and IP (40%)
A third risk is invisibility. Personal AI accounts have no enterprise retention controls, no audit logs, and no way to cut off access when someone leaves. IBM’s VP of Security and Runtime Products, Suja Viswesan, warned that the gap between AI adoption and oversight “already exists, and threat actors are starting to exploit it.”
Here is how sanctioned and shadow AI compare on the controls that matter:
Control | Enterprise AI | Personal / Shadow AI |
|---|---|---|
Data retention | Configurable, can opt out of training | Often feeds training, hard to delete |
Data residency | Known and contracted | Unknown |
Audit logging | Full history of activity | None |
Access at offboarding | Revoked when the person leaves | Account stays with the employee |
Compliance coverage | Covered by BAAs and DPAs | No coverage |
Do not ban AI. Bans push usage underground, where you have even less visibility than before. The better path is controlled enablement: give people good tools, then set clear rules. A practical five-pillar framework looks like this:
Two standards give this structure. The NIST AI Risk Management Framework helps you govern, map, measure, and manage AI risk across its lifecycle. ISO/IEC 42001 sets requirements for an AI management system covering risk assessment, transparency, and accountability. Pairing approved tools with training tends to cut unauthorized use, because most employees reach for shadow AI only when the sanctioned option is missing or weak. Done well, governance protects you and pays off.
Shadow AI is already inside most companies, whether leadership has noticed or not. The fix is not fear or a blanket ban. It is visibility, clear data rules, and tools people actually want to use.
If you are shaping how your business uses AI this year, building a deliberate AI strategy is the place to start. Bliss Drive’s AI visibility team can help you put that plan together.
No. Using AI without approval is not a crime by itself. The problem is what it can trigger. Inputting regulated data into an unvetted tool can violate privacy laws like GDPR, HIPAA, or CCPA, and can void compliance agreements your business depends on. The legal exposure comes from the data, not the tool.
Shadow IT is the unsanctioned use of software or hardware. Shadow AI goes further because external models actively process and often retain the data employees give them. A model can absorb sensitive input and surface it later, which creates a data leakage path that traditional shadow IT does not.
Customer information is the top exposure. IBM found that AI-related incidents involving shadow AI are significantly more likely to expose customer PII (65%) and intellectual property (40%). Source code, client records, financial data, and internal strategy documents are the categories most often pasted into public tools.
No. Bans rarely work and usually move the activity out of sight. A more effective path is to provide approved tools, define what data is off-limits, and train staff on safe use. Visibility plus a good sanctioned option beats prohibition every time.
