Whether Claude is safe for company data depends mostly on the plan. Anthropic's business plans, Team and Enterprise, run under its Commercial Terms, and its 2025 consumer training changes do not apply to them. Personal Free, Pro and Max accounts run under consumer terms. The business still owns the rules about what goes in.
This is general information, not legal advice.
Key takeaways
- The plan type decides how data is treated: business plans use commercial terms, personal plans use consumer terms.
- According to Anthropic, its August 2025 consumer terms changes "do not apply to services under our Commercial Terms", including Team and Enterprise.
- Anthropic's pricing page lists "no training on your content by default" for the Team plan.
- Enterprise adds controls such as audit logs, role-based access and custom data retention.
- Australian privacy law applies to any AI use involving personal information, whichever plan you use.
Is Claude safe for company data?
On a business plan, with sensible rules, it can be used safely for most company work. Business plans run under Anthropic's Commercial Terms, separate from consumer terms. The bigger risks usually come from staff using personal accounts, putting sensitive personal data into any AI tool, and giving connected tools more access than they need.
Safety is not one setting. It combines the vendor's terms, the plan's admin controls, the business's own rules, and how connections to other systems are set up. A good plan does not help if staff paste customer records into a personal account. Equally, careful staff cannot fix a plan with no admin controls.
How do Claude's consumer and business plans differ?
Consumer plans (Free, Pro and Max) run under consumer terms. Since Anthropic's August 2025 update, those users choose whether their chats and coding sessions are used to improve Claude. Business plans (Team and Enterprise) run under the Commercial Terms, which the August 2025 changes did not affect.
According to Anthropic's announcement of 28 August 2025, consumer users who allow their data to be used for training have it retained for five years. Those who do not keep the existing 30-day period. The same announcement states the changes "do not apply to services under our Commercial Terms", which include Claude for Work (Team and Enterprise), Claude Gov, Claude for Education and the API.
| Consumer: Free, Pro, Max | Business: Team, Enterprise | |
|---|---|---|
| Terms | Consumer terms | Commercial Terms |
| Training on chats | User's choice since August 2025 | Not affected by the consumer change |
| Admin | Individual | Central billing, SSO, admin controls |
| Who owns the account | The individual | The company |
What controls do Claude Team and Enterprise add?
Team adds company ownership of accounts, central billing, single sign-on and admin controls over connectors, and Anthropic's pricing page lists no training on your content by default. Enterprise adds controls larger and regulated organisations need: role-based access, automated provisioning, audit logs, a compliance API, custom data retention and IP allowlisting.
According to Anthropic's pricing page in October 2026, Team suits organisations of 2 to 150 people and Enterprise carries a 20-seat minimum. The comparison of Team, Enterprise and Pro covers the differences in more detail.
Memory has admin controls too. According to Anthropic, Enterprise administrators can turn memory off for the whole organisation.
What does the business still need to do?
The business still needs to move staff onto company accounts, write rules about what data may go into AI, keep sensitive personal information out unless there is a clear basis, scope connections to other systems carefully, and have a person approve anything that goes out. Vendor terms cover the vendor; the rest is yours.
A short checklist:
- Company accounts only for company work.
- A one-page policy on what data may go into which tools.
- Sensitive data rules: personal, health and financial information handled with extra care.
- Connections read only to start, with write access granted per action.
- Human approval for anything published, sent or spent.
Breaches tied to unmanaged AI use are real. According to IBM's 2025 Cost of a Data Breach report, one in five organisations studied reported a breach linked to shadow AI. The guide to writing an AI policy includes a template.
What does Australian privacy law mean for using Claude?
The Privacy Act applies to any AI use involving personal information, whichever vendor or plan you use. Australia's privacy regulator recommends as best practice that organisations do not enter personal information, particularly sensitive information, into publicly available generative AI tools, and expects due diligence before adopting AI products.
The Office of the Australian Information Commissioner published this guidance in October 2024 and updated it in January 2025. It also notes that personal information AI generates or infers, including inaccurate content about an identifiable person, is covered by the Australian Privacy Principles. Choosing a business plan helps; it does not remove the obligations.
Skein helps leadership teams choose the plan, set the data rules and roll them out; the AI leadership page covers those decisions. That suits a business with staff on a mix of personal accounts. Organisations handling health or other sensitive data should involve their privacy or legal adviser before setting the rules.
Frequently asked questions
Does Claude use my business data for training?
Anthropic's business plans, Team and Enterprise, run under its Commercial Terms. Its August 2025 consumer terms update, which let Free, Pro and Max users choose whether chats train models, explicitly does not apply to Commercial Terms services. Anthropic's pricing page lists no training on your content by default for Team.
Is Claude Pro safe for work use?
Pro is an individual plan under consumer terms, so it is not designed for company data. Users choose whether their chats are used for training, and the account belongs to the individual. For company work, Anthropic's Team or Enterprise plans, which run under commercial terms, are the better fit.
What is the difference between Claude's consumer and commercial terms?
Consumer terms cover the Free, Pro and Max plans, where users have chosen since August 2025 whether chats are used for training. Commercial Terms cover Team, Enterprise, Gov, Education and the API, and Anthropic stated its consumer training changes do not apply to them.
Is it legal to use Claude with customer data in Australia?
The Privacy Act applies to any AI use involving personal information. Australia's privacy regulator recommends, as best practice, not entering personal or sensitive information into publicly available generative AI tools, and expects due diligence on AI products. Check your obligations with a privacy adviser before using customer data.
How do I make Claude safe for my team?
Put everyone doing company work on a Team or Enterprise plan, write a one-page policy on what data may go in, keep sensitive personal information out unless there is a clear basis, connect tools read only at first, and have a person approve anything that leaves the business.
Related reading
- Anthropic. "Updates to our consumer terms." Anthropic, 28 August 2025. https://www.anthropic.com/news/updates-to-our-consumer-terms
- Anthropic. "Claude pricing." Anthropic, accessed 30 October 2026. https://claude.com/pricing
- Anthropic. "Bringing memory to Claude." Anthropic, 11 September 2025, updated 23 October 2025. https://claude.com/blog/memory
- Office of the Australian Information Commissioner. "Guidance on privacy and the use of commercially available AI products." OAIC, 21 October 2024, updated 17 January 2025. https://www.oaic.gov.au/privacy/privacy-guidance-for-organisations-and-government-agencies/guidance-on-privacy-and-the-use-of-commercially-available-ai-products
- IBM. "IBM Report: 13% of Organizations Reported Breaches of AI Models or Applications." IBM Newsroom, 30 July 2025. https://newsroom.ibm.com/2025-07-30-ibm-report-13-of-organizations-reported-breaches-of-ai-models-or-applications,-97-of-which-reported-lacking-proper-ai-access-controls