What Is a Private AI Workspace?

A private AI workspace is an AI tool that a firm runs for its own people, kept separate from other customers, under business terms that do not let the model provider train on the firm’s data. The firm controls who can sign in, how long conversations are kept, which AI providers are used, and can see who used AI and when. For law and accounting firms, it is the middle ground between staff using personal chatbot accounts and building AI systems in-house.

What makes an AI workspace private?

Five properties, taken together. A tool that has only one or two of them is a business chatbot, not a private workspace.

1. Separation. The firm’s conversations and files are kept apart from other customers. The strongest form is a separate application and database for each firm. A weaker form is one shared system with access rules between accounts.
2. No training. The provider’s terms say it will not train its models on the firm’s prompts, files or answers. This should be the default, not a setting each person must find.
3. Firm control of access. The firm decides who gets in, removes people when they leave, and requires two-factor authentication.
4. Firm control of data. The firm sets how long conversations are kept and which AI providers can receive its data.
5. A record. Partners can see who used AI, when, with which model and at what cost.

How is a private AI workspace different from ChatGPT or another chatbot?

The difference is who holds the controls. In a consumer chatbot account, each person accepts the terms, chooses the settings and owns the history. In a private AI workspace, the firm does.

That matters for the data terms too. On personal plans, OpenAI says it may use your content to train its models unless you turn that off, and Anthropic says it uses consumer chats for training if the person allows it in a Model Improvement setting. On their business products, both say they do not train on customer data by default. Our guides on ChatGPT and Claude set out the details.

A business plan from a model provider, such as ChatGPT Business or Claude Team, already gives a firm several of the five properties. A private AI workspace adds firm-level separation, and often tools built for a profession, such as assistants for contract review or IRS notices.

Is a private AI workspace the same as a private LLM?

No. A private LLM usually means a model the firm hosts itself, on its own servers or a dedicated cloud account, so prompts never go to a model provider. A private AI workspace usually sends requests to a commercial model, such as Claude, through a business account that does not train on the data.

Each has trade-offs:

	Private AI workspace	Self-hosted private LLM
Model quality	Current commercial models	Usually smaller open models
Who receives prompts	The model provider, under business terms	No outside model provider
Who runs it	The vendor	The firm or its IT provider
Cost to start	Per-seat subscription	Hardware or cloud spend, plus setup
Upkeep	The vendor updates it	The firm patches, secures and updates it

For most small and mid-size firms, running their own model is more work and more risk than they want. A well-run workspace with clear provider terms is the practical choice.

What does “private” not mean?

It does not mean nothing ever leaves the firm, and no vendor should tell you otherwise. When a model answers a question, the request goes to the model provider. Providers keep some data for a limited time. For example, Anthropic says it deletes API inputs and outputs within 30 days, and keeps content flagged for a usage policy violation for up to 2 years.

Other things to check:

• Web search. If the tool searches the web, search terms go to search engines.
• Other providers. If the workspace offers several AI providers, each has its own terms and processing locations.
• Legal holds. Any provider may keep data longer if the law requires it.
• The vendor’s own staff. Ask who at the vendor can read your content, and whether you would see it in an audit log.

Why do law and accounting firms need a private AI workspace?

Because their rules ask who else can see client information. ABA Formal Opinion 512 says lawyers must assess the risk that information relating to a representation could be accessed by others before entering it into an AI tool. The AICPA Code says a CPA should have a confidentiality agreement with a third-party service provider, or the client’s specific consent, before sharing confidential client information. Tax preparers also face IRC section 7216.

Courts are paying attention too. In United States v. Heppner, a federal court held that a defendant’s own chats with a consumer AI tool were not privileged, partly because the tool’s privacy policy allowed training and disclosure to third parties. See our guide to Heppner.

A private AI workspace does not settle these duties. It gives the firm the terms and controls it needs to meet them.

How do you evaluate a private AI workspace?

Ask for answers in writing, and check them against the vendor’s legal pages:

1. Is each firm’s data kept in its own application and database, or separated by access rules?
2. Which AI providers receive data, under which terms, and where are they processed?
3. Does any provider train on the data? How long does each keep it?
4. Who at the vendor can read content, and is that logged where the firm can see it?
5. Can the firm set retention, require two-factor sign-in and see a usage record?
6. Which independent audits or certifications does the vendor hold, if any?

Our due-diligence checklist expands each question.

How Zrico fits

Zrico is a private AI workspace for US law and accounting firms. Each firm gets its own application and databases, on a private network no other firm can reach. Claude is the default and is processed only in the US; the firm owner can turn on GPT and Gemini, whose processing is not limited to the US. No AI provider trains on prompts, files or answers. Zrico’s admin tools cannot show chats, files or prompts. The owner gets a usage and audit report and retention settings. Zrico has no SOC 2, ISO or HIPAA certification, and its output must be reviewed by a professional. The security page has the detail.

Sources

• OpenAI Help Center, 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
• OpenAI, Enterprise privacy at OpenAI: https://openai.com/enterprise-privacy/
• Anthropic Privacy Center, Is my data used for model training? (consumer products, March 16, 2026): https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training
• Anthropic Privacy Center, Is my data used for model training? (commercial products, August 18, 2026): https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-training
• Anthropic Privacy Center, How long do you store my organization’s data? (July 1, 2026): https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-data
• ABA Formal Opinion 512, Generative Artificial Intelligence Tools (July 29, 2024): https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf
• AICPA Code of Professional Conduct (see 1.700.001 and 1.700.040): https://pub.aicpa.org/codeofconduct/ethicsresources/et-cod.pdf
• 26 U.S.C. §7216, Disclosure or use of information by preparers of returns: https://www.law.cornell.edu/uscode/text/26/7216
• United States v. Heppner, No. 25 Cr. 503 (JSR), Memorandum, ECF No. 27 (S.D.N.Y. Feb. 17, 2026): https://storage.courtlistener.com/recap/gov.uscourts.nysd.652138/gov.uscourts.nysd.652138.27.0.pdf

General information, not legal advice. Check the current rules in your jurisdiction.