Open source vs proprietary: AI and law

Open source vs proprietary (AI models)

In AI, open-source models (Llama, Mistral, DeepSeek) make their code and weights publicly available and can be hosted on-premises. Proprietary models (GPT, Claude, Gemini) are closed and accessible only through an API. For law firms, open source offers a decisive advantage: fully local hosting guarantees that data never leaves the firm.

The distinction between open-source and proprietary models is a strategic question for legal professionals. Open-source models such as Llama (Meta), Mistral (France) and DeepSeek (China) make their weights and architecture publicly available. They can be downloaded and hosted on a firm's or company's own servers. Proprietary models such as GPT (OpenAI), Claude (Anthropic) and Gemini (Google) are closed: they are accessible only through cloud APIs.

For law firms, the central issue is confidentiality. With an open-source model hosted locally, client data never leaves the firm's infrastructure — no risk of transmission to a third party, no concerns around the US Cloud Act, and full compliance with professional secrecy and attorney-client privilege. This is open source's killer argument for regulated professions.

Mistral, the French AI player valued at around $10 billion in 2025, embodies the digital sovereignty stakes. Its models deliver competitive performance while being deployable in Europe, on GDPR-compliant infrastructure. The choice between open source and proprietary is not, however, binary: many legaltech solutions combine both approaches, using open source for sensitive processing and proprietary APIs for less critical tasks.

Related terms