Have a question? Ask it to contact@cursivetechnology.com
Q: Isn’t all keylogging bad?
A: No.
At Cursive, we securely collect and analyze data from specific text areas only to support learning through assessment. This uniquely valuable dataset brings transparency and new insights to the writing process when collected transparently, from online text areas on institutional learning management systems or places where users have opted-in to collection.
It is true that keylogger programs can be used for malicious purposes: to steal your password or collect sensitive communications.
Below you will find additional context on data collection (we collect only from the active text editor, and only where permitted by the institution, teacher, or writer), and the sensitive nature of the data and its security.
Q: Is this an AI detector?
A: No.
AI detection is the statistical analysis of static text by an algorithm on static text. AI detection tools do serve a purpose and can be useful in many settings (even the classroom).
However, detection tools can be easily beaten with a minimum of effort and uploaded documents and pasted text hold no insights to the student’s thinking or learning process.
At Cursive, we are not solving a problem introduced by ChatGPT or Generative AI, we are solving a problem that became much more visible when these tools entered the arena: no existing tools tied writers to their writing and we were collectively ignoring the value of the writing process. Our focus is behavioral data rather than just relying on the end product or artifact uploaded as a submission.
Q: Can’t students just retype from ChatGPT?
A: No.
Firstly, forcing a would-be cheater to spend nearly the same amount of effort than a student who is completing your assignment honestly significantly raises the bar of academic integrity (the bar to cheat is as low as ctrl+v currently).
Authentic writing processes are messy and complicated, made up of several discrete activities. To quote one of our favorite papers: “authentic writing is a dynamic process where writers monitor their writing and evaluate whether the writing needs to be changed if problems are identified. Transcribed writing, on the other hand is much more linear.1“
In short, you can see it through our playback feature and we can detect it.
Q: how student data is handled?
A: Securely and privately.
From day one, we recognized the sensitivity of this data type. By rule, we do not collect personal identifying information such as email, name, or other data from the LMS or through our Browser Extension. Our mission is to create a zero-trust architecture that helps us serve individuals and institutions in their pursuit of authorship verification. While we verify authorship through biometrics, we ourselves do not need to verify against any particular identity (as long as it’s the same identity).
All policies and technical configurations are designed to minimize risk and maximize privacy and security. This includes TLS 1.2 level security and authentication for all communications, best-in-class security features and standards from our cloud service provider (AWS), and positioning the company and infrastructure to meet the highest security standards. Our HECVAT lite can be accessed from the security policy page. We’re available to answer more technical questions as needed.
Q: How does Cursive know to only record what’s happening on a certain text area and not on a different window that might be private?
A: the logging capacity for our technology is confined to either the text editor within an LMS as part of the end-user configuration of the browser extension.
For the latter, this includes any URLs added by permission (action required) by the end user to the extension, and on those pages is limited to the active text editor field on the loaded page. It does not collect data on actions or activities outside the active text editor.
Q: What languages are supported?
A: Our plugin and app are language agnostic. Since we’re focused on the behavioral data from the process of writing, our ML and analytics will still work for capturing new and relevant information about the writing process, including Authorship. We have tested and continue to make progress on the following languages:
English, French, Spanish, German, Swedish, & Arabic.
We are currently testing to support Chinese, Korean, Japanese (if your institution is interested in a “currently testing” or other language please contact us.
- Plagiarism Detection Using Keystroke Logs https://doi.org/10.5281/zenodo.12729864 ↩︎