Is Claude safe for work?
15 min read
Is Claude safe for work? A dated, sourced table of how Anthropic handles company data on Free, Pro, Max, Team, Enterprise and the API, and a verdict per plan.
It depends on which Claude. Anthropic sells Claude under two sets of terms: consumer terms for Free, Pro, Max and Ultra, and commercial terms for Team, Enterprise and the API. The short answer, as of 15 November 2026:
Plan | Safe for company data? | Why |
|---|---|---|
Free, Pro, Max, Ultra | No | Anthropic lists model training on these plans as "Opt-out". Chats that train are kept for up to five years. No admin, no audit trail, and a shared link is open to anyone who has it. |
Team | Yes, for most work | No training on your content by default, SSO, central billing, sharing limited to your organisation. No SCIM, no audit logs, no retention setting, and chats stay until someone deletes them. |
Enterprise | Yes, if US storage is acceptable | SCIM, audit logs, a Compliance API, custom retention, your own encryption keys, a HIPAA-ready option. Anthropic states that data is stored in the US, and the only location control is US-only inference. |
API | Yes, if you build on it | No training, inputs and outputs deleted within 30 days, zero data retention on approval. Storage is in the US. |
Everything below is drawn from Anthropic'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 The table is the part that goes out of date, so check the review date before you quote it in a policy. Claude Code follows the same split of terms but adds telemetry, permissions and managed settings, so it has its own guide: Is Claude Code safe for work?
What Anthropic does with your data, by plan
Free, Pro, Max, Ultra | Team | Enterprise | API | |
|---|---|---|---|---|
Trains on your content by default | Yes, unless the user opts out. The Consumer Terms allow training "unless you opt out of training through your account settings" | No. "None by default" | No. "None by default" | No. The Commercial Terms say Anthropic "may not train models on Customer Content" |
How to turn training off | Settings › Privacy › Help improve our AI models. Thumbs up or down still sends that whole conversation, and chats flagged for safety review can still be used | Not needed. Owners can switch off the feedback buttons with the Rate chats setting | Not needed. Same Rate chats setting | Not needed. Opt-in only, for example the Development Partner Program |
How long chats are kept | Until the user deletes them. With training on, de-identified copies are kept up to 5 years. Incognito chats 30 days | Until a user deletes them. No retention setting. Incognito chats 30 days | Indefinitely by default. Owners can set a retention period, minimum 30 days, for chats, projects and artifacts | Inputs and outputs deleted within 30 days. Files API content until you delete it |
Deleted chats purged in | 30 days | 30 days | 30 days | Not applicable; automatic within 30 days |
Zero data retention | No | No | Chat and Cowork: no. Claude Code only, for qualified accounts that Anthropic enables | Yes, on approval, for eligible features. Fable and Mythos models need 30-day retention |
Where it is stored | Not stated by region. Anthropic "may process data in different countries where we or our partners operate" | "Data is stored in the US." Traffic may route to the US, Europe, Asia and Australia | Same as Team | Workspace storage: US is the only option. Inference: global by default |
Residency options | None | None | US-only inference on usage-based plans, billed at 1.1x on newer models. It does not change where data is stored |
|
SSO and SCIM | None | SSO, domain verification, just-in-time provisioning. No SCIM | SSO, domain capture, SCIM, role-based access, tenant restrictions, IP allowlisting | SSO and SCIM for Claude Console organisations; tenant restrictions |
Audit trail | None | None. The Primary Owner can export the organisation's conversations | Audit log export covering 180 days, without chat content. Compliance API with an Activity Feed kept for 6 years, plus retrieval and deletion of chats, files and projects | Activity Feed through the Compliance API with an Admin API key |
Other admin controls | Per-user only: incognito chats, memory, a list of shared chats | Central billing, connector controls, usage analytics, spend caps, desktop app deployment, a block on new organisations on your domain | User and organisation spend limits, customer-managed encryption keys, custom roles, HIPAA-ready configuration | Workspaces, Admin API, customer-managed keys, HIPAA readiness with a BAA |
Certifications | Not listed for consumer plans | SOC 2 Type 2, ISO 27001, ISO 42001, CSA STAR, NIST 800-171. HIPAA: not applicable | The same, plus HIPAA | The same, plus HIPAA |
Price | $0; Pro $20 a month, or $17 billed annually; Max $100; Ultra $200 | $25 per seat a month, or $20 annual. Premium seat $125, or $100 annual. 2 to 150 seats | $20 per seat a month, billed annually, plus usage at API rates. 20 seats minimum self-serve, 50 through sales | Usage-based |
Anthropic states that data is encrypted in transit and at rest on every plan. Ultra is the new name for the old Max 20x plan, and Max is the old Max 5x. "Claude for Work" in Anthropic's documents means Team and Enterprise together.
What changed in 2025, and what applies today
On 28 August 2025 Anthropic announced new Consumer Terms and a new privacy policy. It began training new models on chats and coding sessions from Free, Pro and Max accounts where the user allows it. Existing users had until 8 October 2025 to choose, and after that date could not keep using Claude without choosing. For users who allow training, retention went from 30 days to five years. The change covers new or resumed chats only.
The change did not touch anything under the Commercial Terms: Team, Enterprise, the API, Amazon Bedrock and Google Cloud.
Today the Consumer Terms, effective 8 October 2025, say Anthropic may use your material to train models unless you opt out in your account settings. Anthropic's pricing page lists model training as "Opt-out" for Free, Pro, Max and Ultra. Two uses continue after an opt-out: conversations you rate with the thumbs buttons, and conversations flagged for safety review.
What that means for your staff
On a personal account, which is what most employees have, the training setting belongs to the employee. Nobody at your company can see what it is set to, see what was pasted, or delete it. A paid Pro or Max subscription changes the usage limits. It does not change the terms.
On Team, the training risk is gone and sign-in can run through your identity provider. Chat links can only be shared with members of the same organisation. What is missing is the machinery to prove compliance: no audit log, no retention period, no SCIM. Just-in-time provisioning never removes anyone, so a leaver keeps their seat until an admin removes them. Team suits drafting, analysis of non-restricted data and internal code that is not a trade secret.
On Enterprise, the controls match what you run for email and file storage: retention you set, SCIM, an audit feed, roles, your own encryption keys. The gap for a UK or EU organisation is location. Anthropic stores the data in the US and offers no UK or EU storage in its own apps. If the model must run in a UK or EU region, the route is Claude through Amazon Bedrock or Google Cloud. Anthropic's documentation says both offer Claude models in UK and EU regions, and that the cloud provider processes the data in the region you select. Our guide to AI data residency covers that decision.
The red list: what never goes into any Claude
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.
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 the HIPAA-ready configuration and a signed BAA
Security incident details while the incident is open
Anthropic's own help page gives consumers similar advice: be careful with financial details, health records, passwords and confidential business documents.
The risks that stay, whatever plan you buy
Prefilled prompt links. On 18 March 2026 Oasis Security published "Claudy Day". A link to claude.ai could carry a prompt with hidden instructions. If the user submitted it, Claude could send conversation history and memory to an attacker's account through the Files API. An open redirect on claude.com let the link sit behind a search advert. Anthropic fixed the prompt injection; Oasis reported the other issues as still being addressed.
The desktop app. On 15 July 2026 Oasis published "PromptFiction". A claude:// link opened Claude Desktop and submitted an attacker's prompt with one click. On a default install it could send earlier conversations out. Claude Desktop 1.1.2321 and later make the user review the prompt and press send. Anthropic's trust centre also lists CVE-2026-22561, a DLL search order hijacking flaw in the Claude for Windows installer. Desktop AI apps need patching like any other software.
Connectors and prompt injection. A document, web page or email that Claude reads can carry instructions that the model follows. Connectors to Google Drive, Slack, GitHub and Microsoft 365 widen what such an instruction can reach. Team and Enterprise admins can control connectors; personal accounts cannot be controlled.
Retention carve-outs. The 30-day deletion window has exceptions on every plan. Content flagged as a usage policy violation is kept for up to 2 years and its classifier scores for up to 7. A conversation sent with a thumbs rating is kept for 5 years. Anthropic may also keep data where the law requires it.
Shared links on personal accounts. Anyone with the link can read the shared snapshot. Team and Enterprise limit sharing to the organisation.
Wrong answers. Claude states false things fluently, on every plan. The fix is process: 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):
Settings › Privacy › turn off "Help improve our AI models".
Use incognito chats for anything sensitive, and still assume 30 days of retention.
Do not use the thumbs up or down buttons.
Settings › Privacy › Shared chats › Manage, and unshare what should not be public.
Team organisation:
Verify your domain, set up SSO, then turn on "Require SSO for Claude".
Turn on "Restrict organization creation" so staff cannot open new Claude accounts on your domain.
Organization settings › Data and Privacy › turn off "Rate chats".
Decide which connectors are allowed.
Remove leavers by hand.
Write down that there is no audit log and no retention period, and tell staff.
Enterprise organisation:
Set a retention period. Projects are kept indefinitely by default, and a project's setting overrides the chat setting for chats inside it.
Turn on SCIM directory sync so leavers lose access when the identity provider removes them.
Enable the Compliance API and feed it to your SIEM. Recording starts when you enable it and is not backfilled.
Add the tenant restrictions header at your network proxy, so staff on the corporate network cannot sign in to a personal account.
Turn on US-only inference if a contract requires it, and note that it does not move storage.
Add a customer-managed encryption key if you need to be able to withdraw Anthropic's access.
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. 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.
Keep sensitive work on approved models
No Claude plan runs Anthropic's models on a machine you control. For the red list, the choice is a different model behind the same app.
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 and Claude Code are both supported. Once connected, the Claude app sends its requests to the model that Link serves. That is an open-weight model on your hardware; Anthropic's model weights do not run locally. A rule can say: this team's requests run on the device, and if the device cannot serve them, fail. When no allowed destination exists, the request fails with an error. 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:
Claude Team or Enterprise for general work, with the settings above.
A Claude app routed by Aventail™ for anything on the red list: a local model on the employee's own machine, or an approved provider through your own account, with request records in Control that show where each request went.
Two limits to know. While Claude uses the Connect gateway, account-based claude.ai connectors are unavailable. And Connect governs model requests only, so the app's file access, plugins and other network activity keep their own settings.
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 steps are in our guide to Claude Code with Aventail™ Connect.
Frequently asked questions
Does Claude train on my data?
On Free, Pro, Max and Ultra, yes, unless you turn off "Help improve our AI models" under Settings › Privacy. On Team, Enterprise and the API, no by default. On every plan, a conversation you rate with the thumbs buttons can be used.
Is Claude Pro safe for confidential information?
No. Pro is a consumer plan. The training setting is the user's own, there is no admin, and there is no record of what was pasted.
Can my employer see my Claude chats?
On a Team or Enterprise plan, yes. The Primary Owner can export the organisation's conversations, and incognito chats are included. Enterprise organisations can also retrieve chats through the Compliance API. On a personal account your employer sees nothing, which is the problem for the employer.
What is the difference between Claude Team and Enterprise for security?
Both have no training by default and SSO. Enterprise adds SCIM, audit logs, the Compliance API, custom data retention, role-based access, customer-managed encryption keys, IP allowlisting and the HIPAA-ready option.
Does Anthropic store Claude data in the UK or EU?
No. Anthropic states that commercial data is stored in the US, and the API's only storage region is the US. Inference may run in the US, Europe, Asia or Australia unless you choose US-only. For a UK or EU region, use Claude through Amazon Bedrock or Google Cloud.
Is Claude HIPAA compliant?
Anthropic offers a HIPAA-ready configuration with a business associate agreement on Enterprise and on the API. Team and the consumer plans are not covered.
Is Claude safer than ChatGPT for work?
The consumer plans of both train on chats unless the user opts out, and the business plans of both do not. The main difference on the day we checked is location: ChatGPT Enterprise offers storage in the UK and EU, and Claude Enterprise stores data in the US. See Is ChatGPT safe for work?
Related guides
Sources
All read on 15 November 2026.
Anthropic, Updates to Consumer Terms and Privacy Policy (28 August 2025): https://www.anthropic.com/news/updates-to-our-consumer-terms
Anthropic, Consumer Terms of Service (effective 8 October 2025): https://www.anthropic.com/legal/consumer-terms
Anthropic, Commercial Terms of Service: https://www.anthropic.com/legal/commercial-terms
Claude pricing: https://claude.com/pricing
Anthropic Trust Center: https://trust.anthropic.com/
Privacy Center, Is my data used for model training? (consumer): https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training
Privacy Center, Is my data used for model training? (commercial): https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training
Privacy Center, How long do you store my data? (consumer): https://privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data
Privacy Center, How long do you store my organization's data?: https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-data
Privacy Center, How do I change my model improvement privacy settings?: https://privacy.claude.com/en/articles/12109829-how-do-i-change-my-model-improvement-privacy-settings
Privacy Center, Where are your servers located?: https://privacy.claude.com/en/articles/7996890-where-are-your-servers-located-do-you-host-your-models-on-eu-servers
Privacy Center, Configure custom data retention controls for Enterprise plans: https://privacy.claude.com/en/articles/10440198-configure-custom-data-retention-controls-for-enterprise-plans
Privacy Center, I have a zero data retention agreement with Anthropic: https://privacy.claude.com/en/articles/8956058-i-have-a-zero-data-retention-agreement-with-anthropic-what-products-does-it-apply-to
Privacy Center, Data retention practices for Covered Models: https://privacy.claude.com/en/articles/15425996-data-retention-practices-for-covered-models
Privacy Center, What certifications has Anthropic obtained?: https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained
Privacy Center, How does Anthropic protect the personal data of Claude users?: https://privacy.claude.com/en/articles/10458704-how-does-anthropic-protect-the-personal-data-of-claude-users
Privacy Center, Who owns and manages the data of my team?: https://privacy.claude.com/en/articles/9265372-who-owns-and-manages-the-data-of-my-team
Claude Help, What is the Team plan?: https://support.claude.com/en/articles/9266767-what-is-the-team-plan
Claude Help, What is the Enterprise plan?: https://support.claude.com/en/articles/9797531-what-is-the-enterprise-plan
Claude Help, Enable US-only inference for your organization: https://support.claude.com/en/articles/15422948-enable-us-only-inference-for-your-organization
Claude Help, Set up single sign-on (SSO): https://support.claude.com/en/articles/13132885-set-up-single-sign-on-sso
Claude Help, Set up JIT or SCIM provisioning: https://support.claude.com/en/articles/13133195-set-up-jit-or-scim-provisioning
Claude Help, Access audit logs: https://support.claude.com/en/articles/9970975-access-audit-logs
Claude Help, Export your organization's data: https://support.claude.com/en/articles/13346720-export-your-organization-s-data
Claude Help, Use incognito chats: https://support.claude.com/en/articles/12260368-use-incognito-chats
Claude Help, Share and unshare chats: https://support.claude.com/en/articles/10593882-share-and-unshare-chats
Claude Help, Who can view my conversations?: https://support.claude.com/en/articles/8325621-i-would-like-to-input-sensitive-data-into-my-chats-with-claude-who-can-view-my-conversations
Claude Help, Enforce network-level access control with Tenant Restrictions: https://support.claude.com/en/articles/13198485-enforce-network-level-access-control-with-tenant-restrictions
Claude Platform docs, API and data retention: https://platform.claude.com/docs/en/manage-claude/api-and-data-retention
Claude Platform docs, Data residency: https://platform.claude.com/docs/en/manage-claude/data-residency
Claude Platform docs, Compliance API and Activity Feed: https://platform.claude.com/docs/en/manage-claude/compliance-api and https://platform.claude.com/docs/en/manage-claude/compliance-activity-feed
Claude Platform docs, Customer-managed encryption keys: https://platform.claude.com/docs/en/manage-claude/cmek
Claude docs, Claude Desktop on third-party platforms: https://claude.com/docs/third-party/claude-desktop/overview
Claude Code docs, Feature availability (plan names, SSO, SCIM and Compliance API by plan): https://code.claude.com/docs/en/feature-availability
Oasis Security, Claudy Day (18 March 2026, updated 27 May 2026): https://oasis.security/blog/claude-ai-prompt-injection-data-exfiltration-vulnerability
Oasis Security, PromptFiction (15 July 2026, updated 28 July 2026): https://www.oasis.security/blog/claude-desktop-vulnerability
Aventail™, Is ChatGPT safe for work?: https://aventail.co.uk/guides/is-chatgpt-safe-for-work
Aventail™, Connect overview: https://docs.aventail.co.uk/docs/connect/overview
Aventail™, Trust: https://aventail.co.uk/trust
Aventail™, home page plans: https://aventail.co.uk/