As a platform that leverages a global community of data scientists to build its models, the first question we get asked is often: How do you secure clients’ data and IP?
Rest assured that keeping your data secure is our top priority.
Our approach to data protection depends on the sensitivity of data we are handling:
- Public data: Data that exists in some public form is shared openly with our community of Solvers.
- Non-sensitive private data: Data that may not be very sensitive, like sensor data, network data, or resource data, is generally shared after being anonymized and approved by the clients.
- Sensitive private data: Sensitive data is handled in one of the following secure ways:
- Data that the client is comfortable sharing with CrowdANALYTIX but not its community of Solvers is shared via a simulation of the data. Solvers can build baseline models that are tuned by our in-house team.
- Data that the client doesn’t want leaving their private firewalls is handled by deploying our platform behind the client’s firewall so that the data and models all remain only on the client’s servers. Other than performance metrics, no data moves past the firewalls. See the architecture below:
All CrowdANALYTIX community members (known as “Solvers”) sign an agreement with us to transfer IP of all solutions they build and get paid for to CrowdANALYTIX. CrowdANALYTIX in turn takes on any future liability associated with those models. As a result, the liability and the responsibility of protecting the models is that of CrowdANALYTIX. If needed, community members can sign further clauses on a case-by-case basis to handle client-specific requirements.
Our solutions are developed either by customizing pre-built components from our API library, or by building several independent models that are combined into a final AI solution.
In both cases, our Solvers either don’t know the exact problem they are solving, or are exposed to only a small component of the solution. The model that they may have developed is inoperable unless it has access to the expected input data, and the input data that the model relies on is transformed using a script that only CrowdANALYTIX has access to. As a result, no individual Solver or team of Solvers can replicate the solution for any reason.