Explainer
AI data privacy for regulated industries
Where public AI leaks data, and how an owned, in-perimeter model makes privacy structural.
By James Drayson
In short
AI data privacy is about ensuring the data you put into an AI system isn't exposed, retained, or used to train someone else's model. Public AI APIs transmit your data to third parties; an owned model deployed inside your perimeter keeps data private by design, the safest approach for regulated industries.
Where data leaks with public AI
- Transmission: Prompts and documents are sent to the provider's servers.
- Retention: Inputs may be logged or stored under the provider's terms.
- Training reuse: Data can be absorbed into third-party model training unless explicitly excluded.
- Sub-processors: Your data may pass through parties you never chose.
Privacy by design with owned models
- In-perimeter inference: Data never leaves your environment, so there's nothing to transmit or retain externally.
- No third-party training: Your data is never used to improve someone else's model.
- Air-gap option: For the most sensitive data, run with no external connectivity at all.
- Owned & auditable: You hold the weights and logs, supporting privacy assurances.
What this looks like with Locai
Compliance is far easier to evidence when the machine, the model and the logs are all things you own, not things a vendor describes in a whitepaper.
Locai Labs builds Locai One, an on-prem AI appliance. It is one machine that arrives with everything already in it: the hardware, our open-weight Locai Juno models, and Locai OS, the operating system that serves the models and handles users, access and monitoring. You plug it into a mains socket and your network, and your team is working in about 15 minutes. No cloud account, no per-token bill, and nothing leaving the building.
The reason a data-centre-class model fits in a box on your floor is SPACE, our compression algorithm. Instead of asking how much of a model can be cut while keeping it broadly similar, SPACE asks what the model needs to be good at, preserves the subnetworks behind those capabilities and strips back the rest. The result is a smaller specialist rather than a shrunken generalist, tuned to the exact hardware it ships on.
Locai One starts at £29,950 for a team, and Locai One Pro at £49,950 for an organisation, bought once and owned outright. Both run air-cooled on standard mains power and work fully air-gapped. Any compatible open-weight model runs alongside Juno, and if you need a model trained on your own proprietary data we can post-train one and deploy it on the same machine.
Frequently asked questions
Is my data safe with AI?
It depends on the deployment. With a public API your data is sent to a third party; with an owned, in-perimeter model it never leaves your environment, the safest option.
Do AI providers train on my data?
Some may, unless you've explicitly opted out or have contractual exclusions. An owned model removes the question entirely, your data is never shared.
How do I keep AI private?
Run a model you control inside your perimeter, on-prem, in your cloud tenant, or air-gapped, so data is private by design.
What about air-gapped deployment?
Air-gapping fully isolates the model from external networks, giving the strongest privacy guarantee for the most sensitive data.
Book a sovereign AI briefing
A 30-minute session on owning your model: deployment options, the data path, and a clear cost range for your use case.