Shadow AI audit.
Your team already uses AI tools you never approved. We find them, show what data goes into them, and replace them with approved tools that are set up safely, without banning AI.
Where unapproved AI hides
What is shadow AI?
Shadow AI is AI tools and AI features used at work without the company knowing or approving them. The risk isn’t the tool itself. It’s the data that goes into it, and the terms it goes in on.
Unapproved tools
Personal ChatGPT, Claude or Gemini accounts, AI browser extensions, note-taker bots and AI apps on phones, used for work.
Unapproved features
AI switched on inside tools you already pay for, such as a CRM, helpdesk or document editor, without anyone reviewing it.
Approved tools on the wrong terms
A tool the company allows, used on a personal or free plan with different training and retention settings.
The channels we check.
Personal chat accounts are the obvious place. These four are the ones IT teams most often miss.
Browser extensions
What we look forAI writing, summarising and sidebar extensions on managed and unmanaged browsers, and the site access each one holds.
Why it mattersAn extension with access to every site can read whatever is on screen, including CRM records, webmail and internal tools.
AI plugins and connectors
What we look forApps and connectors linked to chat assistants, AI add-ins in office suites, AI apps in Slack or Teams, and OAuth grants to AI apps.
Why it mattersA connector can give an AI tool standing access to mailboxes, drives or tickets, far beyond what one person pastes in.
Meeting recorders
What we look forNote-taker bots that join calls, and recording apps on laptops and phones.
Why it mattersRecordings capture everyone on the call, customers included. Under India’s DPDP Act, a recorded voice of an identifiable person is personal data. See AI meeting recorders and consent.
Coding assistants
What we look forCode assistants in editors and terminals, the plan each developer is on, and API keys paid on personal cards.
Why it mattersPlans differ. As of 20 November 2026, GitHub says Copilot Free, Pro and Pro+ interaction data, including code snippets, is used for training unless the user opts out, while Business and Enterprise are not affected. See the hidden data flows in AI coding assistants.
How we find it, from records you already have.
We count which tools are used and on what terms. We don’t read what individual people typed, and we agree with you how staff are told first. In India, the DPDP Act applies to personal data in AI tools too.
Single sign-on and OAuth grants
Which AI apps people signed into with a work account, and what access those apps were given.
Expense and card data
AI subscriptions paid on company or reimbursed cards.
Browser extension inventory
Extensions installed on managed browsers, and the permissions they hold.
DNS and proxy logs
Traffic to AI services, counted by tool, not read by person.
SaaS admin settings
AI features enabled inside tools you already pay for.
A short staff survey
What people use and why, which also shows the tools they need.
What you get, and what we fix.
The goal isn’t a list of offences. It’s an approved way to use AI that’s easier than the unapproved one.
AI tool register
What we findEvery AI tool and AI feature found, who uses it, and how much.
What we fixA register you keep, with an owner for each approved tool.
Data exposure map
What we findWhich kinds of data go into each tool: customer, employee, code, financial.
What we fixThe highest-risk flows ranked first, so you fix those first.
Terms and settings
What we findConsumer or business plan, training settings, retention and data region.
What we fixApproved tools moved to work accounts with the right settings, recorded.
Safe alternatives
What we findWhat people were trying to get done with each unapproved tool.
What we fixAn approved tool for each real need, set up and easy to reach.
AI use policy
What we findWhat staff may put into which tool, in plain words.
What we fixA one-page policy that matches the approved tools.
Data leakage and PII redaction.
Shadow AI is mostly a data problem, in any country. Our guide to PII redaction for LLMs covers masking before the prompt and restoring after.
Where data leaks
Customer records, contracts, code and staff data pasted into consumer accounts, read by extensions, or held by note-takers on terms nobody checked. Client contracts may also limit where their data can go, wherever you operate.
Redact before the prompt
For AI features and internal tools that call a model’s API, remove or replace personal data before the request leaves your systems, and log what was masked. Open-source detectors such as Presidio include Indian identifiers like PAN, Aadhaar, GSTIN, passport and voter ID numbers.
Test it on your own data
Redaction reduces leakage; it doesn’t remove it. Test how much it catches on your own documents, and mask Aadhaar numbers to the last four digits before any model call.
Pair it with other controls
Approved work accounts with the right settings, access control on what the AI can retrieve, clean logs, and a private model for data that should never leave your infrastructure.
Discovery tool or audit service: what you get from each.
They solve different parts of the problem, and some companies use both.
What it is
A discovery toolSoftware that watches sign-ins, extensions or network traffic for AI apps, on an ongoing basis.
A shadow AI auditA fixed-scope review that finds the AI in use, traces the data, and fixes what it finds.
What it sees
A discovery toolWhat its connectors and agents can reach: usually sign-ins, traffic and installed software.
A shadow AI auditThe same records, plus SaaS admin settings, vendor terms, expense data and what staff tell us in the survey.
Data and terms
A discovery toolFlags the app, sometimes with a risk score.
A shadow AI auditReads the plan, training and retention settings for each tool, and maps which data goes into it.
Fixes
A discovery toolAlerts, nudges or blocks, depending on the product.
A shadow AI auditApproved tools set up on work accounts, risky ones closed, and a policy that matches.
Over time
A discovery toolContinuous, for as long as you run it.
A shadow AI auditA point-in-time baseline, plus a simple way to keep watch afterwards.
Best for
A discovery toolLarge estates that need to spot new tools as they appear.
A shadow AI auditA first baseline, or evidence before an enterprise security review. Many teams start with an audit, then decide if they need a tool.
How it runs.
A shadow AI audit can run on its own, or as part of AI governance work before an enterprise review.
- Step 1
Discover
We pull the records above with your IT team and run the staff survey.
- Step 2
Map
We trace which data reaches each tool, and on what terms.
- Step 3
Replace
We set up approved tools, close the risky ones, and write the policy.
- Step 4
Keep watch
We leave you a simple way to spot new AI tools as they appear.
How long it takes, and what sets the pace.
We agree the scope and timeline with you before we start. These four things decide how long it runs.
How many people and teams
More teams means more tools, more survey answers and more data flows to trace.
Which records are ready
Discovery moves fastest when your IT team can export single sign-on, expense, browser and proxy data at the start.
How much needs replacing
Each unapproved tool with a real need behind it needs an approved alternative set up before it is closed.
Who needs to sign off
Vendor terms, the staff notice and the AI use policy often need legal, HR or leadership approval.
What happens after the audit.
The audit ends with approved AI that people actually use, not a report that sits in a folder.
Approved tools go live
Staff get the approved tools on work accounts, with the settings recorded, before the risky ones are closed.
The policy is published
A one-page AI use policy that matches the approved tools. Our AI acceptable use policy template is a starting point.
The register has an owner
Someone keeps the AI tool register current, with an owner for each approved tool and a simple way for staff to ask for a new one.
You keep watch
Recheck sign-ins, extensions and SaaS AI settings on a regular cycle, and again before a big customer review.
Evidence is ready
The register, settings and policy are the records enterprise buyers ask for in AI security questionnaires.
Related services.
DPDP and AI tools
Where personal data of people in India reaches AI tools, checked against the DPDP Act.
See DPDP and AI toolsAI Adoption
Internal AI tools your team will actually use, so nobody needs a personal account.
See AI AdoptionOwn Your AI
Private models on infrastructure you control, for data that should never leave it.
See Own Your AIShadow AI FAQs.
What is shadow AI?
Shadow AI is AI tools and AI features used at work without the company knowing or approving them: personal ChatGPT accounts, browser extensions, meeting note-takers, or AI features switched on inside existing SaaS tools. The risk is not the tool itself but the data that goes into it and the terms it goes in on.
Should we just ban AI tools?
Bans rarely work: people move to personal phones and accounts, and you lose sight of the data altogether. It is usually safer to approve a few tools on work accounts with the right settings, and make that the easiest option.
How do you find AI tools nobody listed?
From the records you already have: single sign-on and OAuth grants, expense and card data, browser extension inventories, DNS or proxy logs, and the AI settings inside your existing SaaS tools. A short staff survey fills the gaps. We look at which tools are used and on what terms, not at what individual people typed.
How does this relate to DPDP compliance?
If your product handles personal data of people in India, the DPDP Act covers the personal data that reaches your AI features and AI vendors too, with most of its duties starting on 13 May 2027. We cover that overlap in our DPDP work, and the same evidence often answers both your buyers and the Act.
Is this legal advice?
No. We work on the engineering and the evidence: systems, settings, logs and documents. Contract terms and legal questions go to your counsel, or to our partner law firm if you want us to involve them.