What Is the Difference Between Public AI and Private AI? Why It Matters for Your Data

Public AI refers to AI services available to anyone who creates an account, typically on a subscription or freemium basis. ChatGPT at chat.openai.com, Claude at claude.ai, and Google Gemini at gemini.google.com are public AI services.

Using these services means sending your inputs to the AI provider's infrastructure and receiving outputs from that infrastructure. The data governance questions are: how does the provider use the data you submit? Is it used to train future models? How long is it retained? Where is it processed? Who can access it?

For consumer use cases, the providers' terms of service typically provide adequate protections. For business use cases involving confidential information, client data, personal data, or commercially sensitive material, the default consumer terms of service are almost certainly insufficient and may directly breach your contractual and regulatory obligations.

Private AI refers to AI services deployed within a controlled environment with specific contractual data protection commitments. The clearest example is Microsoft Azure OpenAI: Microsoft makes the same OpenAI models available within Azure, but under enterprise terms where Microsoft commits that customer data is not used to train models, data is processed within the customer's selected Azure regions, and the service is subject to the same data protection agreements as other Azure services.

Microsoft 365 Copilot is similarly a private AI deployment: it uses foundation models but processes your data within your Microsoft 365 tenant, governed by Microsoft's enterprise data protection commitments.

Claude enterprise API deployments and Google Vertex AI provide analogous private deployment options for their respective models. The pattern is consistent: enterprise-tier access to leading AI models with data protection commitments appropriate for business use.

The governance problem in most organisations is that employees discover the productivity benefits of public AI tools before policies are in place, and informal use of public AI with business data becomes widespread.

This creates concrete risk. Personal data submitted to a public AI tool may trigger UK GDPR obligations if the provider is not a contracted data processor. Client confidential information submitted to a public AI tool may breach professional obligations, client contracts, or NDAs. Commercially sensitive information may be exposed in ways that affect IP protection or competitive position.

CIOs and GCs have found themselves dealing with data breach notifications, contract disputes, and regulatory enquiries arising from employee use of public AI tools with business data. The solution is not to ban AI use, which rarely works, but to provide governed alternatives that deliver the productivity benefits employees are seeking within appropriate data governance controls.

Boards should ensure that their organisations have a clear policy distinguishing between approved enterprise AI tools (private AI with appropriate data governance) and consumer tools (public AI not suitable for business data).

This policy needs to: clearly specify which tools are approved and for what types of data; provide employees with enterprise AI tools that meet their productivity needs; and include communication and training so that employees understand why the distinction matters.

The policy also needs to address the grey zone: tools that started as consumer services but now offer enterprise tiers with improved data governance. Many organisations have found that providing approved enterprise AI substantially reduces shadow AI use, because employees were using public tools out of necessity rather than preference.