Skip to content
Aventail™ investigation into UK government AI dependence featured in Computer Weekly.Read the story
Aventail™
All guides

Is ChatGPT safe for work?

13 min read

Is ChatGPT safe for work? A dated, sourced table of what OpenAI does with company data on each plan, a verdict per plan, and the settings to change.

It depends on which ChatGPT. The name covers five products with different data terms. The short answer, as of 15 November 2026:

Plan

Safe for company data?

Why

Free, Plus, Pro

No

Chats train OpenAI's models unless each person turns it off. No admin, no retention control, no record of what was pasted. Treat every chat as public.

Business

Yes, for most work

No training on your content, SSO and domain verification, admins can view and delete member chats. No SCIM, no compliance log feed, shared links cannot be switched off, no zero-retention option.

Enterprise and Edu

Yes

Admin-set retention, data residency in the UK or EU, SSO and SCIM, a compliance API, role-based permissions, key management. The product IT expected all along.

API

Yes, with approval

No training by default, 30-day abuse logs, zero data retention on approval for eligible endpoints. Only relevant if you build on it.

Everything below is drawn from OpenAI's own pages, read on 15 November 2026 and linked at the end. The table is the part that goes out of date, so check the review date before you quote it in a policy.

What OpenAI does with your data, by plan

Free, Plus, Pro

Business

Enterprise and Edu

API

Trains on your content by default

Yes. "Data sharing is enabled by default" on personal workspaces

No. Only a separate signed agreement can enable it

No

No, since March 2023

How to turn training off

Settings › Data controls › Improve the model for everyone, or the Privacy Portal. New chats only. Thumbs up or down on a reply still lets that whole conversation be used

Not needed. Admins can also switch off response feedback

Not needed

Not needed; opt-in only

How long chats are kept

Until the user deletes them. Temporary chats up to 30 days

Admins control retention. Content of removed members is retained indefinitely

Admins set it: indefinite or time-bound, such as 90 or 180 days

Abuse-monitoring logs up to 30 days; uploaded files until deleted

Deleted chats purged in

30 days, unless de-identified or on legal hold

30 days

30 days

30 days

Zero data retention

No

No. A "contracted offering", not self-serve

No public terms

Yes, on approval, for eligible endpoints

Where it is stored

"Various jurisdictions, including the United States"

Region chosen at checkout, rolling out. Abuse-monitoring copies and a time-limited copy of every prompt stay in the US

Region set when the workspace is created

Global by default; regional projects by endpoint

Residency options

None

Storage only, not inference

Storage at rest in the UK, EEA and Switzerland, Australia, Canada, India, Japan, Singapore, South Korea, UAE, US. Inference residency in the EU, US and UAE. No extra cost

Storage in 10 regions; in-region processing in the US and EU, limited elsewhere. 10% price uplift on newer models

SSO and SCIM

None

SAML and OIDC SSO, domain verification. No SCIM, no synced groups

SSO, SCIM (Okta, Entra ID, Google Workspace, Ping, OneLogin, Rippling), up to 99 verified domains

SSO and SCIM through an Enterprise tenant

Audit trail

None

None. Admins can view, export and delete member chats by hand

Compliance Logs Platform and Compliance API, 30-day feed, integrations with Purview, Netskope, Zscaler, Palo Alto, CrowdStrike, Varonis, Smarsh and others

Organisation audit logs via the Admin API

Other admin controls

Per-user only: memory, temporary chat, shared links

Roles (owner, admin, analytics viewer, member), spend controls, connectors to Google Workspace, Slack, GitHub, Microsoft 365. Members can invite others; shared links cannot be disabled workspace-wide

Permissions per model, GPT, Canvas, Projects, memory, link sharing; apps and custom MCP on or off per role; IP allowlisting; Lockdown Mode; Enterprise Key Management; Intune

Project-level residency, Modified Abuse Monitoring, customer keys via AWS KMS, Google Cloud KMS or Azure Key Vault

Certifications

Not listed per product

SOC 2 Type 2; ISO 27001 not listed for Business

SOC 2 Type 2, ISO 27001:2022, ISO 27701:2019; HIPAA BAA with Regulated Workspace

SOC 2 Type 2, ISO 27001, ISO 27701; BAA with modified retention

Price

$0; $20 a month; $100 to $500 a month

$25 per user a month, or $20 annual; Premium $125 or $100. Usually two seats minimum, 200 maximum; card only

Contact sales; invoicing available

Usage-based

Encryption is the same everywhere OpenAI states it: AES-256 at rest, TLS 1.2 or higher in transit. ChatGPT Team was renamed ChatGPT Business on 29 August 2025, so an older policy that names Team means Business.

What that means for your staff

On a personal account, which is what most employees have, anything they paste is training data by default and sits on OpenAI's servers until they delete it. The thumbs-up button quietly re-enables training for that conversation. Nobody at your company can see what was pasted, delete it, or prove it was not pasted. That is why the Samsung engineers who uploaded source code in April 2023 triggered a company-wide ban within a month.

On Business, the training risk is gone and sign-in runs through your identity provider. What is missing is the machinery to prove compliance: there is no log feed, retention of a leaver's content is indefinite, and a member can share a link to a chat with anyone because the workspace cannot turn sharing off. Business is fine for drafting, summarising, analysis of non-restricted data and internal code that is not a trade secret.

On Enterprise, the controls match what you already run for email and file storage: retention you set, data at rest in the UK, a compliance feed into your existing tooling, roles, and a kill switch per feature. The remaining risks are the ones no vendor setting removes, covered below.

The red list: what never goes into any ChatGPT

Put this list in the policy and in the onboarding deck. It applies on every plan, Enterprise included, because the risk is the person pasting it, not the vendor training on it.

  • Passwords, API keys, tokens, connection strings, private keys

  • Customer personal data: names with contact details, account numbers, health or financial records

  • Source code that is a trade secret, or anything under a customer NDA

  • Unreleased financial results, deal terms, board papers

  • Anything legally privileged

  • Patient or health data, unless you are on Enterprise with a signed BAA and the Regulated Workspace

  • Security incident details while the incident is open

If staff need an AI tool for work that touches this list, the answer is not "be careful". It is a tool that cannot send the data out, which is the last section.

The risks that stay, whatever plan you buy

Shared links. On 26 October 2026 OpenAI fixed a flaw where opening a shared ChatGPT link ran attacker code inside the viewer's signed-in session and read a private conversation. Opening the link was the only action required. Business workspaces cannot disable link sharing; Enterprise can.

The desktop app. On 18 September 2026 OpenAI fixed a bypass of Lockdown Mode in the ChatGPT desktop app that allowed zero-click data exfiltration through HTML preview popups. Lockdown Mode had been introduced in June 2026. Treat desktop AI apps as software that needs patching, because it is.

Legal holds. From May to September 2025 a US court order in the New York Times case made OpenAI keep every consumer, Plus, Pro, Team and non-ZDR API conversation, deleted or not. Enterprise and Edu were excluded, as were UK, EEA and Swiss users. The order ended on 26 September 2025, but OpenAI still holds the April to September 2025 data. A 30-day deletion promise is a policy, and a court can suspend a policy.

The 2023 history bug. For nine hours on 20 March 2023 some users saw the titles of other users' chats, and 1.2% of active Plus subscribers had payment details exposed. It was a library bug, not an attack. Multi-tenant software has these.

Prompt injection. A document, web page or email that ChatGPT reads can carry instructions that the model follows. Connectors to Google Drive, Slack and GitHub widen what an injected instruction can reach. Enterprise lets you switch connectors off per role; Business does not.

Wrong answers. ChatGPT states false things fluently. Every plan. The fix is process, not settings: nothing goes to a customer, a regulator or production without a human who understands it.

Settings to change today

Every personal account used for work (until you move people off them):

  1. Settings › Data controls › turn off "Improve the model for everyone".

  2. Turn on two-factor authentication.

  3. Use temporary chats for anything sensitive, and still assume 30 days of retention.

  4. Do not use the thumbs up or down buttons.

Business workspace:

  1. Turn on SAML or OIDC SSO and verify your domain, so personal accounts on your domain are pulled into the workspace.

  2. Require MFA at the identity provider.

  3. Admin Console › turn off "Allow response feedback".

  4. Set roles: one or two owners, the rest members.

  5. Write down that shared links are on and cannot be turned off, and tell staff.

  6. Decide what happens to a leaver's chats, because by default they are kept indefinitely.

Enterprise workspace:

  1. Set a retention period. 90 or 180 days is where most start.

  2. Pick the storage region at provisioning; it cannot be moved later. For UK data, choose the UK.

  3. Turn on SCIM so leavers lose access the day they leave.

  4. Connect the Compliance API to the DLP or SIEM you already run.

  5. Set permissions per role: which models, whether GPTs can be shared outside, which apps and connectors are on.

  6. Turn on Lockdown Mode and IP allowlisting for roles that handle restricted data.

What to put in the acceptable use policy

Six lines cover it. Which plan is approved and which accounts are banned for work. The red list above. Who may turn on connectors and apps. That outputs are checked before they leave the building. Where to report a paste that should not have happened, without blame, so you hear about it. And the tool to use instead for work on the red list.

Our guide to shadow AI and acceptable use policies has a full template and the numbers on how many staff use unapproved tools today.

Keep sensitive work on approved models

ChatGPT cannot be pointed at your own models. Nothing in Business or Enterprise lets a prompt run on a machine you control. So the question for the red list is not "which ChatGPT plan" but "which tool".

Aventail™ Connect configures supported AI apps to send their model requests through Aventail™ Link, an inference engine and gateway that runs on each machine, under routing rules set in Aventail Control. Claude Desktop, Claude Code, OpenCode and Codex are supported today. A rule can say: this team's requests run on the device; if the device cannot serve them, fail, do not send them anywhere else. When no allowed destination exists, the request fails with an error. It is not sent to a disallowed destination. Prompts and responses do not pass through the hosted Control service; it receives management, device and routing events.

In practice that gives you two lanes:

  • ChatGPT Business or Enterprise for general work, with the settings above.

  • An approved assistant routed by Aventail™ for anything on the red list: a local model on the developer's own machine, or an approved provider through your own account, with a Requests view in Control that shows where each request went.

Setup is the Connect button in Link, with no endpoint or key to enter. The free plan covers three connected devices, including business use. The step-by-step for developers is in our guide to Claude Code with a local model.

Frequently asked questions

Does ChatGPT Business keep data private?
It does not train on your content, and OpenAI states it holds it under the Business terms with encryption at rest and in transit. Admins can see, export and delete member chats. It is private from OpenAI's models, not from your own admins, and shared links are still on.

Can others see my chats in ChatGPT Business?
Workspace owners and admins can view and export a member's conversations. Colleagues cannot, unless you share a link or a GPT with them.

Is ChatGPT safe for confidential information?
On Free, Plus or Pro, no. On Business, only information you would be comfortable having retained indefinitely and visible to admins. On Enterprise, within the retention and region you set. Information on the red list above does not go into ChatGPT on any plan.

Is ChatGPT Plus safe for confidential information?
No. Plus is a personal product. Training is on by default, there is no admin, and there is no record of what was pasted.

Is ChatGPT safe from hackers?
OpenAI's infrastructure is audited to SOC 2 Type 2 and ISO 27001. The incidents that have hit users were a library bug in 2023, a shared-link flaw fixed in October 2026 and a desktop app bypass fixed in September 2026. Your bigger exposure is a staff member's reused password with no MFA, and phishing pages that imitate ChatGPT.

Is it bad to use ChatGPT for work emails?
Drafting an email is fine on a Business or Enterprise account. Pasting a customer's email thread into a personal account is a data transfer to a third party that your privacy notice probably does not cover.

Does OpenAI store ChatGPT data in the UK?
Enterprise and Edu can choose UK storage at rest at no extra cost. Business residency is storage-only and still rolling out, with abuse-monitoring copies kept in the US. Consumer plans have no residency option.

Sources

All read on 15 November 2026.

  1. OpenAI, Enterprise privacy: https://openai.com/enterprise-privacy/

  2. OpenAI Help, How your data is used to improve model performance: https://help.openai.com/en/articles/5722486-how-your-data-is-used-to-improve-model-performance

  3. OpenAI Help, Data controls in ChatGPT: https://help.openai.com/en/articles/7730893-data-controls-in-chatgpt

  4. ChatGPT pricing: https://chatgpt.com/pricing/

  5. OpenAI Help, Data residency and inference residency for ChatGPT: https://help.openai.com/en/articles/9903489-data-residency-and-inference-residency-for-chatgpt

  6. OpenAI Help, Where your ChatGPT Business content is stored: https://help.openai.com/en/articles/20001418-where-your-chatgpt-business-content-is-stored

  7. OpenAI, Your data (API): https://developers.openai.com/api/docs/guides/your-data

  8. OpenAI, Data residency (API): https://developers.openai.com/api/docs/guides/data-residency

  9. OpenAI Trust Portal: https://trust.openai.com/

  10. OpenAI, Product compliance status: https://openai.com/product-compliance-status/

  11. OpenAI, Privacy policy: https://openai.com/policies/privacy-policy/

  12. OpenAI Help, Chat and file retention in ChatGPT: https://help.openai.com/en/articles/8983778-chat-and-file-retention-in-chatgpt

  13. OpenAI Help, Data retention when a member is removed from a workspace: https://help.openai.com/en/articles/8266418-data-retention-when-a-member-is-removed-from-a-workspace

  14. OpenAI Help, Compliance Platform for Enterprise and Edu: https://help.openai.com/en/articles/9261474-openai-compliance-platform-for-enterprise-and-edu-customers

  15. OpenAI Help, ChatGPT Enterprise admin quickstart: https://help.openai.com/en/articles/20001264-chatgpt-enterprise-admin-quickstart

  16. OpenAI Help, Managing workspace settings in ChatGPT Enterprise: https://help.openai.com/en/articles/8411955-managing-workspace-settings-in-chatgpt-enterprise

  17. OpenAI Help, ChatGPT Business overview: https://help.openai.com/en/articles/8792828-chatgpt-business-overview

  18. OpenAI Help, Managing billing and seats in ChatGPT Business: https://help.openai.com/en/articles/8792536-managing-billing-and-seats-in-chatgpt-business

  19. OpenAI Help, Managing data sharing and privacy in ChatGPT Business: https://help.openai.com/en/articles/8798634-managing-data-sharing-and-privacy-in-chatgpt-business

  20. OpenAI Help, Setting up SSO for ChatGPT Business: https://help.openai.com/en/articles/11489188-setting-up-single-sign-on-sso-for-chatgpt-business

  21. OpenAI Help, SCIM provisioning and management: https://help.openai.com/en/articles/10011769-scim-provisioning-and-management

  22. OpenAI Help, HIPAA-eligible products and functionality: https://help.openai.com/en/articles/20001069-hipaa-eligible-products-and-functionality

  23. OpenAI Help, What is ChatGPT Plus: https://help.openai.com/en/articles/6950777-what-is-chatgpt-plus

  24. OpenAI Help, About ChatGPT Pro tiers: https://help.openai.com/en/articles/9793128-about-chatgpt-pro-tiers

  25. OpenAI, March 20 ChatGPT outage: https://openai.com/index/march-20-chatgpt-outage/

  26. Bloomberg via Yahoo Finance, Samsung bans generative AI (2 May 2023): https://ca.finance.yahoo.com/news/samsung-bans-generative-ai-staff-004831399.html

  27. OpenAI, Response to NYT data demands (5 June 2025, updated 22 October 2025): https://openai.com/index/response-to-nyt-data-demands/

  28. Engadget, OpenAI no longer has to preserve all of its ChatGPT data (11 October 2025): https://www.engadget.com/ai/openai-no-longer-has-to-preserve-all-of-its-chatgpt-data-with-some-exceptions-192422093.html

  29. Parameter, We found a gap at OpenAI (28 October 2026): https://www.parameter.ai/labs/we-found-a-gap-at-openai

  30. The Hacker News, Researchers bypass ChatGPT Lockdown Mode (5 November 2026): https://thehackernews.com/2026/11/researchers-bypass-chatgpt-lockdown.html

  31. Aventail™, Connect overview: https://docs.aventail.co.uk/docs/connect/overview

  32. Aventail™, Trust: https://aventail.co.uk/trust

We use optional cookies and tracking to identify visiting companies, load our contact form, measure Google Ads download clicks and understand page use with PostHog on our homepage, Connect and How it works pages, Ollama and LM Studio comparison guides and campaign pages. Cookie policy