Shadow AI control is security software that finds out which outside AI tools a company's employees use, and controls what they may type, paste or upload into them. Shadow AI means AI tools that employees use for work without the IT team's approval, such as a free chatbot opened in a browser. The control is usually a feature of the security gateway that a company already uses for its employees' web traffic.
Why does a company care? Whatever an employee pastes into an outside chatbot leaves the company. The company that runs the chatbot may store it, and on some free plans may use it to train its models. If the text includes customers' personal data, the company is still responsible for that data under India's DPDP Act.
Here is how the control works. Every company laptop sends its web traffic through a secure web gateway, a security cloud that filters the web, wherever the laptop is. This is part of a SASE service. The gateway keeps a list of thousands of AI apps, each with a risk rating. For each app, the security team picks a rule. It can block the app, allow it, allow it only with a company account, or allow it but check what is pasted. The checks reuse the company's data loss prevention (DLP) rules, which recognise things such as customer IDs, source code and documents marked confidential. Instead of blocking, the gateway can also show a warning that explains the rule and points to an approved tool.
Consider a pharmaceutical company in Ahmedabad. A scientist in its research team wants a short summary of a 40-page trial report. She opens a free chatbot and pastes three pages, including patients' ages and the code name of a new compound. The gateway sees text being sent to a chatbot website. Its DLP rule recognises the "Confidential" label and the patient details. It blocks the paste. It shows her a message that points to the company's approved AI assistant, which runs under a contract that keeps the data out of model training. At the end of the month, the security team's report shows 61 different AI tools in use across the company, and 9 of them blocked.
Two neighbouring tools help. A cloud access security broker (CASB) watches what employees do inside approved cloud apps, such as uploading files. SaaS security posture management (SSPM) checks the settings of approved AI apps, such as which company systems each one may read.
The idea behind all of this is zero trust: every request is checked, whether the laptop is in the office or at home. How is this different from LLM guardrails? Guardrails protect an AI application that the company built. Shadow AI control protects the company's data when its employees use AI applications that other companies built. Requests that the company's own programs send to model providers usually pass through an LLM gateway instead.