--- title: Private ChatGPT for business: the options description: Private ChatGPT for business: plans, the API with zero retention, Azure or Bedrock, open models you host, or a private AI service, and who sees your data. canonical: https://privatesuperintelligence.si/blog/private-chatgpt-for-business/ last-updated: 2026-10-15 --- # Private ChatGPT for business: the options A "private ChatGPT" for a business usually means one of five things: a ChatGPT Business or Enterprise workspace, OpenAI's API with zero data retention, OpenAI models hosted inside Microsoft Azure or Amazon Bedrock, an open-weight model you run on your own servers, or a private AI service built for your company. Each step down that list moves your data further from OpenAI's systems, and each one asks more of you or of a vendor. None of these options is wrong. They answer different versions of the question your general counsel or COO is asking: who, outside the firm, can read what we type? ## Five ways to get a private ChatGPT | Option | Where your prompts are processed | Training on your data | Retention you should expect | Who outside your firm can access it | |---|---|---|---|---| | ChatGPT Business or Enterprise | OpenAI | No, by default | Admins set it; deleted chats removed within 30 days | Authorized OpenAI staff in defined cases; on Business, also contractors reviewing for abuse | | OpenAI API with zero data retention | OpenAI | No, by default | None on eligible endpoints, after OpenAI approves you | Limited; some endpoints stay ineligible | | OpenAI models in Azure or Bedrock | Microsoft or AWS, in your cloud account | No | Abuse monitoring by default; can be reduced or turned off | The cloud provider's reviewers in defined cases; not OpenAI | | Open-weight model you host | Your servers or your cloud | You decide | You decide | Whoever you give access to | | Private AI service built for one client | Set per client | Set by contract | Set by contract | Set by contract and design | ## ChatGPT Business and Enterprise This is the shortest path. OpenAI's enterprise privacy page (updated January 8, 2026) says it does not train on data from ChatGPT Business, Enterprise, Edu or the API by default, that you own your inputs and outputs where the law allows, and that data is encrypted at rest with AES-256 and in transit with TLS 1.2 or higher. Business, Enterprise and the API have completed SOC 2 Type 2 audits, and OpenAI will sign a Data Processing Addendum. The access rules differ by tier. On Enterprise, authorized OpenAI employees access conversations only to resolve incidents, to recover conversations with your permission, or where law requires. On Business, authorized employees can access conversations for engineering support, abuse investigation and legal compliance, and so can third-party contractors reviewing for abuse. In both, deleted conversations leave OpenAI's systems within 30 days unless the law requires longer. This is a private workspace, not a private instance. Your chats still sit on OpenAI's servers, which is why a court order in the New York Times case could reach ChatGPT logs. Our guide [Is ChatGPT safe for confidential information?](/blog/is-chatgpt-safe-for-confidential-information/) covers that case, and [ChatGPT Enterprise for family offices](/blog/chatgpt-enterprise-for-family-offices/) looks at the fit for a family office. ## The OpenAI API with zero data retention If your developers build their own chat tool on OpenAI's API, the defaults are stricter. OpenAI says API data is not used for training unless you opt in, and abuse monitoring logs are kept for up to 30 days. OpenAI also offers Modified Abuse Monitoring and Zero Data Retention to approved customers. Under ZDR, OpenAI forces the `store` parameter to false. The catch is coverage. Chat Completions, Responses and Embeddings are eligible. Assistants, Threads and Vector Stores are not, and they keep data until you delete it. A ZDR agreement protects you only if your developers build on the eligible endpoints. Our guide to [zero data retention AI](/blog/zero-data-retention-ai/) lists the exceptions at each provider. ## OpenAI models inside Azure or Amazon Bedrock This is what most people mean by a "private ChatGPT instance." You run OpenAI's models inside a cloud you already contract with. **Microsoft Azure (Foundry Models, including Azure OpenAI).** Microsoft says your prompts and completions are not available to other customers or to OpenAI, are not used to improve OpenAI's models, and that the hosted models do not interact with any service OpenAI operates. Prompts are processed in the geography you choose, unless you pick a Global or Data Zone deployment. By default, an abuse monitoring system can flag prompts for review by authorized Microsoft employees. Approved customers can apply for modified abuse monitoring, and you can check it is off: the resource shows `ContentLogging` set to `false`. **Amazon Bedrock.** AWS added OpenAI's open-weight gpt-oss models in August 2025, and on June 11, 2026 announced GPT-5.4 and GPT-5.5 in its US East (N. Virginia) Region. AWS says model providers have no access to the accounts that run their models on Bedrock, so they cannot see Bedrock logs, prompts or completions. You can set `data_retention_mode` to `none` so that no request or response data is written to durable storage. Some models require AWS human review as a condition of access and are unavailable under that setting, so check each model's `allowed_modes` before you commit. You get the model you know with the contract terms of your cloud provider. You also take on the build: someone has to set up identity, logging and a chat interface, and keep them running. ## Open-weight models you host yourself At the far end, you run the model on hardware you control. OpenAI's gpt-oss-120b is published under the Apache 2.0 license and, by OpenAI's model card, runs on a single 80 GB GPU such as an NVIDIA H100. The smaller gpt-oss-20b fits in 16 GB of memory. Nobody outside your firm sees a prompt unless you let them. The trade is ownership of everything else: serving, updates, security patches, access control and evaluation. Open-weight models are also not the same models as the newest ChatGPT. Our guide on [private LLMs versus a private AI service](/blog/private-llm-vs-private-ai-service/) lays out the staff and work this takes. ## A private AI service built for your company The fifth option hands the build to a provider who designs the system around one client: the hardware, the model provider and where the data lives. You get the privacy of a self-hosted model without staffing the team, and the contract sets the retention and access rules. Our comparison of [building versus buying private AI](/blog/build-vs-buy-private-ai/) walks through that decision. ## Questions to ask whichever option you choose - Who outside the firm can read a prompt, and under what conditions? - How long are prompts, outputs and files kept, and who sets that period? - Which features fall outside the privacy terms (assistants, file stores, web search, connectors)? - Where are prompts processed, and where is stored data kept? - Can you verify the setting yourself, as with Azure's `ContentLogging` flag or Bedrock's retention mode? - What happens to your data if a court orders the provider to preserve it? ## Private SuperIntelligence Private SuperIntelligence from Mitosis Labs is the most capable AI, trusted with everything: an AI concierge for enterprises, family offices and private clients. Each deployment is built for one client, from the hardware to the choice of model provider to where the data is stored, and delivery takes weeks. It works across your own email, documents and data, answers with citations, and acts within permissions you set. [Register interest](/#waitlist) for the next intake. ## Frequently asked questions ### Is there a private version of ChatGPT for business? Yes. ChatGPT Business and Enterprise give you a workspace OpenAI does not train on by default, with admin-set retention. For a separate instance, companies run OpenAI models through Azure or Amazon Bedrock, or host an open-weight model themselves. ### Is ChatGPT Enterprise private? OpenAI does not train on Enterprise data by default, admins control retention, and OpenAI employees access conversations only for incidents, permitted recovery or legal requirements. The data still resides on OpenAI's systems and stays subject to legal process. ### Can I run ChatGPT on my own servers? You can't run the ChatGPT models themselves on your own hardware. You can run OpenAI's open-weight gpt-oss models, published under Apache 2.0, and the larger one runs on a single 80 GB GPU. ### Is Azure OpenAI more private than ChatGPT Enterprise? Microsoft says OpenAI cannot see prompts sent to Azure-hosted models, and that the hosted models don't touch any service OpenAI runs. Microsoft's own abuse monitoring still applies unless you are approved to modify it, so the difference is which company holds your data, and on what terms. ### What does a private ChatGPT cost? It depends on the route: per-seat plans, per-token cloud billing, your own hardware, or a scoped engagement. Ask each vendor for terms in writing alongside the price. ## Related reading - [Is ChatGPT safe for confidential information?](/blog/is-chatgpt-safe-for-confidential-information/) - [What is zero data retention AI?](/blog/zero-data-retention-ai/) - [How to keep company data private when using AI](/blog/keep-company-data-private-with-ai/) - [Private AI vs public AI](/blog/private-ai-vs-public-ai/) - [Is Claude, Copilot or Gemini safe for confidential information?](/blog/is-claude-copilot-gemini-safe-for-confidential-information/) - [What is confidential AI?](/blog/confidential-ai/) - [What is private superintelligence?](/blog/what-is-private-superintelligence/) ## Sources - [Enterprise privacy at OpenAI (OpenAI, updated January 8, 2026)](https://openai.com/enterprise-privacy/), accessed 2026-10-15 - [Data controls in the OpenAI platform (OpenAI)](https://developers.openai.com/api/docs/guides/your-data), accessed 2026-10-15 - [Data, privacy, and security for Foundry Models sold by Azure (Microsoft Learn)](https://learn.microsoft.com/en-us/azure/ai-foundry/responsible-ai/openai/data-privacy), accessed 2026-10-15 - [Data protection (Amazon Bedrock User Guide, AWS)](https://docs.aws.amazon.com/bedrock/latest/userguide/data-protection.html), accessed 2026-10-15 - [Data retention (Amazon Bedrock User Guide, AWS)](https://docs.aws.amazon.com/bedrock/latest/userguide/data-retention.html), accessed 2026-10-15 - [OpenAI open weight models now in Amazon Bedrock and Amazon SageMaker JumpStart (AWS, August 2025)](https://aws.amazon.com/about-aws/whats-new/2025/08/openai-amazon-bedrock-sagemaker-jumpstart/), accessed 2026-10-15 - [OpenAI GPT-5.4 and GPT-5.5 models now available in US East (N. Virginia) on Amazon Bedrock (AWS, June 11, 2026)](https://aws.amazon.com/about-aws/whats-new/2026/06/openai-gpt-us-east-virginia-amazon/), accessed 2026-10-15 - [gpt-oss-120b model card (OpenAI on Hugging Face)](https://huggingface.co/openai/gpt-oss-120b), accessed 2026-10-15