AI tools at work: is each one safe for company data?
6 min read
AI tools at work, one dated guide per tool: ChatGPT, Claude, Copilot, Gemini, DeepSeek and more. A verdict per plan, plus shadow AI and evaluation guides.
Whether an AI tool is safe for company data depends on the plan as much as the product. The same tool can train on what staff paste on a personal account and hold it under contract on an enterprise one. This page links one dated guide for each AI tool people use at work, so a security lead can check a tool in a few minutes and an employee can see which account to use.
Every tool, with its guide
Each guide gives a verdict per plan. The table summarises each verdict as of 15 November 2026. The sources for each one are in its guide.
Tool | Guide | Verdict |
|---|---|---|
ChatGPT | Verdict, 15 November 2026. Free, Plus and Pro: no. Business: yes, for most work. Enterprise and Edu: yes. API: yes, with approval | |
Claude | Verdict, 15 November 2026. Free, Pro, Max and Ultra: no. Team: yes, for most work. Enterprise: yes, if US storage is acceptable. API: yes, if you build on it | |
Copilot | Verdict, 15 November 2026. Copilot on a personal Microsoft account: no. Copilot Chat on a work account: yes, for most work. Microsoft 365 Copilot: yes, after a permissions clean-up. GitHub Copilot is rated by plan in the guide | |
Gemini | Verdict, 15 November 2026. Gemini app on a personal account: no. Gemini in Workspace on a qualifying edition: yes, for most work. API on paid quota: yes, with configuration | |
DeepSeek | Verdict, 15 November 2026. DeepSeek app and API: no. Open weights on your own hardware: yes, after testing | |
Otter AI | Is Otter AI safe for work? | Verdict, 15 November 2026. Basic and Pro: no. Business: only for low-risk internal meetings. Enterprise: yes, once configured |
Grammarly | Is Grammarly safe for work? | Verdict, 15 November 2026. Free: no. Pro: yes for everyday writing, after one change. Enterprise: yes |
Perplexity | Is Perplexity safe for work? | Verdict, 15 November 2026. Free, Pro and Max: no. Enterprise Pro: yes, for most work. Enterprise Max: yes. API: yes, with approval |
NotebookLM | Is NotebookLM safe for work? | Verdict, 15 November 2026. Personal Google account: no. Workspace core service: yes, for most work. Gemini Notebook Enterprise: yes. Google renamed NotebookLM to Gemini Notebook on 16 July 2026 |
Cursor | Is Cursor safe for work? | Verdict, 15 November 2026. Individual plans: no. Teams: yes, for most code, once Privacy Mode is enforced. Enterprise: yes |
Codex | Is Codex safe for work? | Verdict, 15 November 2026. ChatGPT Free, Go, Plus or Pro: no. Business: yes, for most code. Enterprise or Edu: yes. API key: yes, with approval |
Claude Code | Verdict, 15 November 2026. Pro, Max or Ultra subscription: no. Team or Enterprise seat: yes. API key: yes, with device policy. Amazon Bedrock or Google Cloud: yes |
How to read a guide
Every guide in the series has the same parts, in the same order:
A verdict per plan, in a table at the top.
A dated table of what the vendor does with your data: training, retention, where it is stored, sign-on, audit trail, admin controls, certifications and price.
The risks that stay on every plan.
The settings to change today.
Numbered sources, with the date they were read.
The data table is the part that goes out of date. Check the review date before you quote a guide in a policy.
Policy and evaluation
Three guides cover the questions that sit above any single tool.
Shadow AI and AI acceptable use policies. How many staff use unapproved tools, what it costs, how to find it, red, amber and green data classes, and a policy outline to copy.
Our security team has blocked public AI tools. What should we evaluate? Four options compared, named products for each, and a twelve-point checklist.
AI data residency. How to decide where each AI request may run and how to check that it did.
One more guide in this hub compares two models: Gemini 4 Argon vs Carbon.
Keep restricted work on approved models
ChatGPT Business and Enterprise settle training and admin control. Neither lets a prompt run on a machine you control. For work that may not go to any outside vendor, Aventail™ Connect configures supported apps (Claude Desktop, Claude Code, OpenCode and Codex) to send their model requests through Aventail™ Link, which runs on each machine and applies routing rules set in Aventail Control. If a rule permits only the device and the device cannot serve the request, the request fails with an error. It is not sent to a disallowed destination. Aventail™ governs requests routed through it and does not discover or block other tools. Setup is in Claude Code with Aventail™ Connect.
Frequently asked questions
Which AI tools are safe to use at work?
It depends on the plan. For ChatGPT, the personal plans are not safe for company data, Business is for most work, and Enterprise and Edu are. Each guide above gives the answer per plan for its tool.
Is the free version of an AI tool safe for work?
For ChatGPT, no. Chats on personal plans can train OpenAI's models unless each person turns that off, and no admin can see or delete what was pasted.
What should never go into an AI tool?
Passwords and keys, customer personal data, trade-secret source code, unreleased financial results, legally privileged material and details of an open security incident. This applies on every plan.
What if the tool we use is not listed?
Use the twelve-point checklist in the blocked-tools guide. It works for any AI product.
How current are the guides?
Each one shows the date its sources were last read. This page was last reviewed on 15 November 2026.
Related guides
Sources
All read on 15 November 2026.
Aventail™, Is ChatGPT safe for work? The verdict and its OpenAI sources: https://aventail.co.uk/guides/is-chatgpt-safe-for-work
Aventail™, Connect overview: https://docs.aventail.co.uk/docs/connect/overview
Aventail™, Routing: https://docs.aventail.co.uk/docs/fleet/routing
Aventail™, home page (scope of controls): https://aventail.co.uk/