Built so it cannot become a leak between labs.
AI labs avoid shared environment providers for a good reason. If one vendor runs environments for several labs, it could see how each lab's agents behave and pass that knowledge on. This is called distillation risk. benned is designed around removing it.
Distillation risk is the risk that a provider running environments for several labs learns how one lab's models behave and passes that knowledge on. benned removes it structurally rather than contractually: every lab runs on its own physically separate hardware, observations, actions and trajectories are never stored by benned, access keys are held by the lab, and the game licence runs directly between the lab and the studio — so no catalogue ever aggregates several labs' activity in one place.
- 1
Physically separate hardware
Every lab gets its own servers in its own locked rack or cage. No shared GPUs, storage or network between labs.
- 2
No agent data kept
Observations, actions and trajectories stream to the lab and are not stored by benned. We keep usage metering: instance hours and concurrency.
- 3
Keys held by the lab
Access to a lab's environment is granted and revoked by that lab. benned operators get access only for support, on request, and it is logged.
- 4
Direct contracts with studios
Labs license from the studio. benned is not a catalogue that aggregates many labs' activity in one place.
- 5
Hosted in the EU
Dedicated hardware in a Tier III facility in the Netherlands, for labs that want training infrastructure outside other jurisdictions. Deployment inside a lab's own environment is possible.
The same separation protects your game.
Your build runs only on controlled hardware, never on a lab's own machines unless you agree. Each copy is licensed to one lab, for the use you approved.
Ready for your security review.
We share our architecture, access logs and operating procedures with every customer's security team before a pilot starts.
Request a pilot