The second tier of AI assistants raises sharper legal questions than the first, because the differences between them are larger.
Perplexity
Consumer tiers retain data for model improvement by default, with a setting to turn it off; enterprise customers are contractually excluded from training. Perplexity is also a defendant in publisher copyright suits and in Amazon’s case over its shopping agent.
DeepSeek
Its policy lists training as a purpose and says data is stored in the People’s Republic of China, with an opt-out only by request. Italy’s data protection authority ordered it to stop processing Italians’ data in January 2025, several other European authorities opened inquiries, and many US government bodies banned it on official devices.
Mistral Le Chat
The French company began using inputs from its free and paid consumer and team tiers for training by default in 2026, with an opt-out; enterprise is excluded. Hosting in the EU does not by itself answer the training question.
Meta AI
Since December 2025 Meta has used interactions with its AI to personalise content and ads in most regions, with the EU, UK and South Korea excluded pending regulatory clearance.
Self-hosted open-weight models
Running an open-weight model on your own infrastructure means prompts need not leave your systems. The licence terms and security of the deployment then become the main questions.
For confidential work, where the model runs can matter more than which model it is.
What to do
Treat each tool as a supplier: check data location, training use, retention and any open regulatory actions before approving it for work use.
This article is general information on the law as at November 2026. It is not legal advice and has not been updated since that date. See the terms of use.