Sell AI Insights, Not Secrets

Use Duality to personalize and monetize AI while keeping model IP and data safe.

Many organizations today have models of varying types that are valuable when personalized for specific organizations. The challenge is that customer datasets may be too sensitive to send, as are the models themselves. Duality’s Secure Collaborative AI solution eliminates these issues, allowing teams to sell insights, not secrets.

How do I protect my model IP while personalizing models?

Model personalization is when data from a particular organization or population is used to retrain or fine tune an existing model. This traditionally requires either the data being sent to the model owner, or the model being sent to a data owner–often a blocking requirement. By utilizing security and privacy technologies, Duality’s solution provides technical guardrails to ensure that the model IP and input data are protected from view by anyone but the owners. The software features governance controls, reports, and data privacy and security guarantees for both the model and data owners.

How do I prove value to clients?

Show them. Once configured, the encrypted results are returned to the Analyzing Party (model owner). These results can then be shared with the client or the development team for further customization and review.

How it Works

1. Participants pick the location suitable for them and agree upon the key management system to be used.
2. Duality is deployed with a Trusted Execution Environment (TEE) and at each data owner’s and analyzer’s location.
3. All data and models are encrypted onsite before being sent to the TEE to run.
4. The data and models are decrypted inside the TEE, and the computation runs.
5. The results are encrypted and sent back to the analyzing party, where the results can be decrypted.
   Note* No administrator nor provider has access to the data and models inside the TEE. No trust, no problem.

Multi-Cloud Support

Support for this workflow is ubiquitous. It doesn’t matter if the data and model owners are on separate infrastructures or using different data marketplaces. The only common factor will be that each participant is utilizing the same key management system. If your model is hosted on GCP but your client has data in AWS or MS, no problem.

Derek Wood