How do I stop staff pasting client data into ChatGPT? Combine an approved tool, a short rule and technical controls. A ban alone does not hold.
Short answer. Combine four things: a short written rule, an approved AI tool that is good enough for people to actually use, technical controls that warn or block when sensitive data is pasted into unapproved sites, and logs that show what is happening. Banning alone rarely works, because Microsoft and LinkedIn’s 2024 survey found that 78% of AI users bring their own AI tools to work. The controls already exist in common IT tools, for example endpoint data loss prevention that can audit, warn or block a paste according to the destination website, and web filtering that blocks unsanctioned AI sites. Add training and a no-blame way to report slips, and review the rule regularly.
01
Six steps
Write the rule on one page: which AI tools are approved, which data may never be pasted (client identifiers, health records, unpublished financials, tender or examination papers), and who to ask.
Provide an approved tool that is good enough. People route around tools that are slow or weak, which is how shadow AI starts.
Find out what is in use now. Web-filter and proxy logs show which AI sites staff already visit; Microsoft’s shadow AI guide starts with discovering AI apps.
Block unapproved AI sites for the groups that handle client records, and warn or block sensitive pastes into the rest.
Keep logs of approved AI use and review a sample, so the rule is enforced by evidence rather than trust.
Train with real examples and offer a no-blame route to report a slip, so mistakes are reported within hours instead of being hidden.
02
Technical controls, by layer
Other vendors sell comparable features. These are the ones documented by Microsoft, which many offices already license.
Layer
What it does
Documented example
Network or web filtering
Blocks or warns on unapproved AI sites for chosen groups
Microsoft’s prevent-data-leak-to-shadow-AI guide: discover AI apps, then block unsanctioned ones
Endpoint data loss prevention
Audits, warns or blocks pasting sensitive content into a browser according to the destination site
Microsoft Purview “Paste to supported browsers” with sensitive service domain groups; Edge, Chrome, Firefox and Safari supported
Browser prompt inspection
Blocks sensitive text typed or pasted into consumer AI prompts
Microsoft Purview browser data security in Edge covers consumer ChatGPT, Copilot, DeepSeek and Gemini
Encryption and labels
Encrypted, labelled files cannot be processed by AI apps that cannot decrypt them
Microsoft Purview sensitivity labels
Managed workspace
Moves users onto an AI workspace with single sign-on and admin visibility
ChatGPT Business and Enterprise list SAML SSO; Enterprise adds domain verification and SCIM
03
What the survey and the law say
Microsoft and LinkedIn’s 2024 Work Trend Index surveyed 31,000 people in 31 countries: 78% of AI users bring their own tools to work, 80% at small and medium-sized companies, and 52% are reluctant to admit using AI for their most important tasks. A rule that only forbids gives people a reason to stay quiet about a slip.
The Digital Personal Data Protection Act, 2023 defines a personal data breach to include accidental disclosure that compromises confidentiality. When the Rules’ breach provisions apply, from about May 2027, a data fiduciary must tell the Board without delay and give detailed information within 72 hours, and the Act’s penalty schedule allows up to ₹200 crore for failing to give breach notice. A leak nobody notices is therefore a larger problem than one reported early. Whether a given paste is a breach is a legal question for counsel.
04
Where a private system fits
The approved tool can be a managed workspace or a private system inside the office. ANVE logs every model call with the user, app, model and whether it ran locally or in the cloud, so an office can show what the approved tool was used for. Cloud models, connectors and agent tools each need an administrator’s approval, and calls to them are logged. It does not block other websites; that is the job of your firewall or endpoint tools. ANVE is pre-release.
05
Questions people ask
Can we block ChatGPT entirely?
You can block it at the network or browser level on managed devices, but personal phones and home networks are outside that. Pair the block with an approved alternative and a written rule.
Does endpoint data loss prevention cover personal devices?
Microsoft’s Purview endpoint DLP monitors activity on managed devices. Staff phones and personal laptops need a policy and an approved tool instead.
What counts as sensitive data in the rule?
Name the categories in plain words: client names and identifiers, patient records, unpublished financials, tender documents, examination papers. Use the same list in training and in any DLP patterns.
Is a private AI system required?
No. What is required is an approved tool that people can use. A managed workspace may do if your clients accept cloud processing; a private system fits where they do not.
06
What is not real yet
No sealed ANVE unit exists; every image of the appliance on this site is a labelled render. The software runs today on a development machine. No customers yet, no certifications, no delivery date, and the final price is not announced.
Reservations are ₹999 in India or US$12 elsewhere, refundable in full until you confirm an order in writing (terms). Offices that cannot use cloud AI can apply to be one of ten design partners.
