Your email client now has an AI that can summarize every thread you’ve ever been part of. Your video call platform records meetings and generates a transcript before the call ends. Your document editor suggests edits based on context from your other files.
This is not a future scenario. If your company uses Microsoft 365 or Google Workspace, this is already active — or your IT team is deciding whether to turn it on.
Gemini and Copilot are genuinely useful. But before you start using them, or your company enables them without much announcement, it is worth understanding three things: what these tools can see, where that data goes, and where the real risks sit.
The short answer: The enterprise versions of Gemini and Copilot carry documented commitments. Your company’s data stays in its own tenant and is not used to train Google’s or Microsoft’s foundation models. The consumer versions do not carry those commitments. The real risk is not the enterprise tool. It is “shadow AI”: staff pasting work data into personal ChatGPT or Gemini accounts, which sends it outside the company entirely. That is what happened at Samsung in 2023.
What Copilot can access inside Microsoft 365
Microsoft Copilot for Microsoft 365 — the paid enterprise add-on, distinct from the consumer Copilot built into Windows — plugs into your entire Microsoft environment. That means your Outlook inbox and sent mail, documents stored in SharePoint and OneDrive, Teams meeting recordings and transcripts, your calendar, and your chat history.
Say you’re a manager who has been at your company for several years. Copilot can, in theory, summarize years of email chains, cross-reference your calendar against your Teams conversations, and pull context from documents you haven’t opened in months. That’s the feature. It’s also the surface area worth thinking about.
Microsoft’s documented position on the enterprise version is clear: your organization’s data does not leave your Microsoft 365 tenant, and it is not used to train Microsoft’s foundation models. The processing happens within your existing data residency boundary — so if your company’s data is configured to stay in the EU or in a specific region, Copilot respects that.
The consumer version — the free Copilot on Bing and Windows — operates under different terms. Conversations there may be reviewed by Microsoft. The enterprise product and the consumer product are not the same thing, and confusing them is a common mistake.
What Gemini can access inside Google Workspace
Google Workspace’s Gemini integration works similarly. When your company pays for Gemini for Workspace, Google’s documented position is that your customer data is not used to train Google’s AI models. The AI processes your content — drafts in Docs, threads in Gmail, files in Drive — but Google states this happens without your data feeding back into model training.
The important distinction is between your work account and a personal Gmail account. If you’re using a personal Google account and have Gemini active in Gmail, your interactions may be retained and used to improve Google’s services, subject to your account settings. The consumer product and the enterprise product carry different commitments.
Google Meet with Gemini AI summaries is worth noting specifically. When this feature is enabled, every meeting in your organization gets a transcript processed by Google’s AI. That includes calls where nobody hit “record.” If your team has back-to-back client calls, that is a significant volume of content being processed. Know whether this feature is on before you assume your calls are not being transcribed.
The actual risk: shadow AI
Here’s where the real exposure tends to happen. It is less about what Microsoft or Google does with enterprise data. It is more about what your employees do when the enterprise tools aren’t available.
When someone can’t access Copilot or Gemini through their work account, they often use their personal accounts instead. A developer pastes a block of unreleased source code into ChatGPT to debug it. A finance analyst uploads a revenue forecast to a free AI tool to build a slide. An HR manager drops a draft termination letter into Gemini on their personal Gmail to clean up the language.
This is what happened at Samsung in 2023. Engineers pasted internal code into ChatGPT — a consumer product — and that data left Samsung’s environment entirely. The incident is often cited incorrectly as a Copilot problem. It was not. It was a shadow AI problem: employees using personal AI tools for work tasks because no sanctioned alternative was available or accessible.
A 2023 Cyberhaven study found that 11% of data employees paste into AI tools is confidential — customer records, source code, regulated financial information. The enterprise AI products from Microsoft and Google are specifically designed to prevent this class of exposure. But they only work if employees actually use them instead of personal accounts.
What this means for Indian organizations
Indian IT services companies, software startups, and enterprise teams using Google Workspace or Microsoft 365 are operating the same tools as their global counterparts, but with additional regulatory obligations.
The DPDPA requires organizations to have appropriate contractual safeguards in place when personal data is shared with third-party processors. An enterprise contract with Microsoft or Google for their business AI products typically satisfies this requirement — these are documented, auditable agreements. A personal ChatGPT or Gemini account used for work data does not.
For companies handling client data on behalf of banks, insurance firms, or healthcare providers — common in Indian IT services — this distinction matters significantly. Processing that data through a consumer AI product, even unintentionally, may breach the client contract and DPDPA obligations simultaneously.
If your organization does not yet have an AI usage policy, that gap is worth flagging to your IT and legal teams. The tools are already in use. The policy needs to catch up.
What you should do as an individual
Know whether AI features are active in your work tools. In Teams, check whether Intelligent Recap and meeting transcription are enabled. In Google Meet, check whether Gemini AI summaries are on. These settings are usually controlled at the admin level, but you can see whether they’re active and confirm that meeting participants know before you start a call.
Understand your company’s AI policy. Most organizations of any size now have one. If yours does not, ask IT or your manager. The policy tells you which tools are sanctioned and which aren’t.
Never paste the following into a consumer AI product: customer records, unreleased source code, financial projections, HR documents, legal correspondence, or anything marked confidential by your organization. If you need AI assistance with this content, use the enterprise tool your company has provisioned, or ask whether one can be set up.
For research and general work: using consumer AI with dummy data, publicly available information, or fully anonymized content carries minimal risk. The line is between public and work-specific, not between AI and no AI.
The enterprise versions of these tools are built on one assumption: businesses need AI processing without training-data exposure. That commitment only holds when you’re actually inside the enterprise product. The moment work data enters a personal account, the boundary disappears.
If your organisation is trying to close that gap, KnytX is building a business plan for team-wide data-exposure monitoring. It isn’t live yet; join the waitlist to be notified at launch.
What you can do right now
Want to see what data brokers expose about you? See what a scan finds → shows an example report: the kinds of sites that list your name, phone, and address. Once KnytX Shield launches it will find and remove these automatically. Join the waitlist to get in early.