July 28, 2023
Mike Olsen, Founder & CEO July 10, 2023
Proctorio is a proctoring software that uses some free, open-source software (like every software company) in our products. As per license agreements, we not only bundle acknowledgments with our software but also publish it publicly here.
Proctorio uses AI to replace traditional proctors in testing centers. The first step is to use a webcam to track the presence of a face. Ten years ago, OpenCV was chosen for this feature because it was the most lightweight framework available for HAAR-based features and Canny edge detection. Although the current version of OpenCV can do much more, Proctorio only uses these two features.
A claim was previously made that Proctorio doesn’t properly license OpenCV, purported by Satheesan and then reiterated by RTL Nieuws in an email to us last month. Fortunately, webarchive clearly refutes this and we can confirm not only has it always been available on our website but is also bundled in the license file (included with every copy of the software) since at least 2014.
One of Proctorio's key features is its end-to-end encryption. This means that image processing must take place on the candidate's device, not in the cloud. Proctorio includes the default models with its own models. We have made it a priority to improve the accuracy of our models, which we have done several times. Measuring the accuracy of our models against the default models is a metric we track by running the images against each other. While the default models might have bias in them, the Proctorio models are trained specifically using webcam images and use diverse datasets by breaking them down into equal groups based on perceived gender, skin tone, and age.
For this to work, the models need to be included with the software and are found directly in the product. Per Google Chrome Web Store guidelines, these have always been bundled with our Chrome browser extension and not hosted somewhere else.
Any claim that they are 7z files bundled with the extension or hosted on a CDN are either using honeytrap files or outdated models obtained from discredited sources.
If you're reading something online about how some researchers tested our models, be careful as they are most certainly not ones we use today with our 5 million active weekly users. You can verify yourself by checking any links these authors claim to have obtained them from. The fake (honeytrap) files have also been removed from the browser extension to further prove this point (and save you some bandwidth on the download!).
Note: This blog post was updated for clarity on October 17, 2023.