Request Screening Inside the Company
Between you and the service there may be a program run by your company. A request from the chat window reaches it first. It reads the text, looks for company data in it, and decides whether it goes to the model.
It is installed for one reason: so that company data does not leave through an employee's inattention.
Where the screening sits
The check can sit right in the browser and see the text you type or paste into the chat window. If it contains data the company has forbidden to send, the request does not go out. Screening like this can be turned on for ChatGPT, Gemini, DeepSeek, and consumer Copilot.
On the work computer. The same check is installed as a separate program, and then it sees more than the browser.
In the company network. All traffic goes through a gateway, and it inspects requests to chat services along with everything else.
What they look for
Personal data of clients and employees, financial and medical information, source code, passwords, and access keys.
One maker of such gateways analyzed anonymized data from its customers between October 2024 and October 2025.
Employees at those companies regularly tried to send data prohibited by policy into AI services: an average of 223 times a month per organization. Only companies that already had such screening in place were counted.
More than half of those cases involved data protected by law: personal, financial, and medical. Source code and credentials came next.
What happens when it triggers
The message is blocked and the request never reaches the model.
There is also a soft mode: a warning with the option to continue. Then the decision stays with you, and the event stays in the log.
The security team sees the trigger. For them it is a routine event, not a report against you.
Why an employee should know this
Screening is not installed everywhere, and you cannot see it from your side. So the rule about company data stays with you: your company may have a gateway, or it may not.
A block on sending is not a service failure. Your company's rule fired. There is no need to work around it through a personal account or a phone: this is exactly the case the rule was written for.
If the blocked task is work you actually need to do, go to your manager. They know what is allowed for your role and which services the company has under contract.
The screening is run by a third party's software, and it recognizes company data with models of its own. It keeps that data out of public chat services, and it reads every one of your requests in full.
Requests that triggered the check may be stored along with the data that was found. Whether they are stored, where, and who reads them is decided by the company when it configures the system.