CogniVowUnderstand the movement.
Study the approach to a click: timing, velocity, corrections, and path shape. The prototype evaluates 30 behavioral features in the browser.
HUMAN OVERSIGHT FOR AN AGENTIC WORLD
Keep the
human in
the decision.
Agents can act on our behalf.
Trust starts with knowing when a person is really in the loop.
Independent research in behavioral biometrics & digital trust.
01 / THE HUMAN SIGNAL EXPLORING TRUST
Natural movement has a story.
Illustrative paths · not a live classification
↗ A person behind the approval⌁ Behavior as a signal◌ Evidence before promises
THE IDEA BEHIND COGNIVOW
Trust shouldn’t
stop at login.
A valid credential can open a session.
It doesn’t tell the whole story of what happens next.
CogniVow is developing a research-led approach to digital trust: exploring how natural interaction patterns can support ongoing authentication and help assess whether a person took part in a specific approval.
Our mission is to help organizations keep meaningful human oversight as software becomes more autonomous. We start with mouse dynamics, test against automation, and publish what works alongside what still needs work.
A LAYERED APPROACH
One click.
More context.
Our research prototype combines three checks. Each adds context; none guarantees human presence.
BEHAVIORAL MODEL⌁
Check the interaction.
Look for a recorded approach and inspect where the click landed. These are explainable heuristics, with real false-positive limits.
Ask a fresh question.
Change the approval target for every attempt. A human-looking recording is not enough if it doesn’t answer the current challenge.
THE RESEARCH, IN YOUR HANDS
Make a move.
See the signal.
Try the actual browser-based model from the Human vs Agent project. Compare your interaction with a simulated instant agent click or a stale recording.
This is a research experiment, not a production security check. A passing score does not prove someone is human or has given informed consent.
Read the findings and limits ↗APPROVAL SANDBOXNO REAL TRANSACTION
ILLUSTRATIVE AGENT REQUEST
Approve an agent’s next step
The agent proposes. You review.
Loading the research model…
Starting the mouse demo records pointer movement only inside the challenge area. Samples are processed in this tab and cleared when you close the challenge. No sample upload or analytics is implemented.
Mouse / trackpad experiment. Keyboard and touch users can explore the simulations; this model does not verify those input methods.
THE RESEARCH NOTEBOOK
Built on experiments.
Open about the edges.
Two connected research directions. One commitment to reporting evidence in context.
01 / HUMAN OVERSIGHT
AGENT→HUMAN?→ACTION
EVIDENCE ≠ PROOF↗
Was a person actually in the loop?
What mouse dynamics can tell us about an approval—and what independent attackers taught us about the limits.
Explore Human vs Agent ↗02 / CONTINUOUS AUTHENTICATION
REAL SIGNAL / SYNTHETIC DATA↗
Can synthetic movement improve authentication?
The TimeGAN study behind our interest in behavioral biometrics, limited enrollment data, and honest model evaluation.
Read the paper summary ↗OUR RESEARCH PRINCIPLE
Earn trust.
Don’t just claim it.
Behavioral evidence can support a decision. Stronger assurance needs more: context, accessible alternatives, and trusted authentication. That’s the future we’re working toward.
See the honest limits ↗