Skip to content
How it works

Four layers that protect the model itself.

Your model runs on partner, processor and customer servers and cloud, on your own servers and plants, and across your edge fleet, lines and field devices. Four independent layers protect it. Each deployment uses the layers it needs.

The mechanism, in plain words

Locked, Shielded, Hidden and Encrypted results: modular defence in depth, configured per deployment.

Hover, tap or focus a step for the explanation.

Read the full mechanism

Locked: the model file is encrypted and tied to the device it is deployed to, so a copied file is built not to run elsewhere. Shielded: while the model runs, shielding on the host works against inspection, debugging, memory dumps, tampering and code injection. Hidden: the weights are mathematically transformed, so a stolen copy never shows the real weights. Encrypted results: where enabled, only authorised readers can see the outputs.

Why four independent layers

Each layer stands on its own, and each makes an attack harder in a different way. A deployment can use one, several or all four, depending on the hardware and the risks. The full threat model and key custody are walked through in the architecture review.

Where it runs

  • Partner and customerServers and cloud
  • Your own estateServers and plants
  • EdgeFleets, lines and field devices
  • Hardware fitAssessed in the architecture review

The four layers

  • Locked
  • Shielded
  • Hidden
  • Encrypted results
Integration

Model formats, target hardware and where protection sits in your build and delivery pipeline are mapped in the architecture review.

Licence, own cloud, enclave or layered protection

Licence contract

Terms on paper.

Read the detail

A licence sets who may use the model and for how long. Once the model is delivered, protection depends on the other party keeping the terms and on your ability to enforce them.

Own cloud or API

The model stays with you.

Read the detail

Serving the model from your own cloud keeps it in house. It does not help when the customer or partner needs the model running on their servers, in their plant or on a device.

Hardware enclave

A protected area of specific chips.

Read the detail

An enclave runs code inside a protected area of specific supported chips. What it protects, and against whom, depends on the chip vendor's design.

Wodan layered protection

Protection built around the model itself.

Read the detail

Four independent layers protect the model file, the running process, the weights and the results, configured to each deployment and confirmed on your hardware in the pilot.

See the full comparison

Questions engineers ask

What does it cost in performance?

Zero loss of accuracy. Very limited performance impact. Confirmed on your model and your hardware in the pilot, before you commit.

Which models can you protect?

Any model, including LLMs. Which layers apply is scoped with our engineers and tested on your model in the pilot.

Our model sits inside an application. What is covered?

We protect the model inside it. Tell us where your IP actually lives, in the trained weights or in the logic around them, and we will tell you honestly what we can cover.

Are the layers steps I must take in order?

No. The four layers are independent. Each deployment uses the layers its hardware and risks call for.

What does post-quantum mean here?

Post-quantum cryptography is encryption designed to stay secure even against future quantum computers. We use it for encrypted results, so outputs recorded today stay unreadable to unauthorised parties later.

Does this stop someone copying the model by querying it?

Copying a model from its inputs and outputs is called distillation. Encrypted results stop unauthorised parties collecting readable outputs. Someone authorised to read the results can still collect them.

We already have a licence contract. Do we replace it?

No. A licence sets the terms. Wodan protects the model itself. They work side by side.

Bring one model and one deployment.

We map which layers fit your hardware and risks, and what a paid pilot would confirm.

Our engineers review your model and deployment context with you.