Instructor LoginRequest Early Access
Built for the AI classroomThe audit trail for AI-era assignments

You shouldn't have to guess what your students did with AI.

In an AI era, the final student submission tells you less than ever. UnProctor gives instructors a complete audit trail — from first draft through every AI interaction to final response — so you can evaluate the learning, not just the output.

UnProctor
ETH 210 — Midterm47:22 remaining
Question 2 of 3

How does Zuboff's concept of “behavioral surplus” challenge traditional notions of privacy, and what are its implications for democratic governance?

⚡ AI UnlockedIndependent draft: 187 words
AI Assistant
YOU
Can you explain what behavioral surplus means?
UNPROCTOR AI
Behavioral surplus refers to the data collected beyond what's needed for the service — the excess that gets sold as predictions about future behavior...
⚠ Flagged: Response contained direct answer to the question prompt.
Ask a follow-up question...

Instructors are making judgment calls
with no data to back them up.

The submission hasn't changed. But how it was produced has. And right now, you have no way to know.

📋 Before AI

The submission was the work.

Instructors could reasonably trust that a written submission reflected a student's genuine engagement with the material. It wasn't perfect, but the assumption was defensible.

  • Essay = student's reasoning and synthesis
  • Outliers were visible — plagiarism detection worked
  • Grading reflected what the student knew
⚡ Now

The submission might be 10% student, 90% AI — and no one knows.

AI detection tools generate false positives. Plagiarism checkers can't catch AI-paraphrased content. Instructors are left to make high-stakes decisions on instinct.

  • No visibility into the writing process
  • Detection tools flag innocent students
  • Policy violations are impossible to prove fairly
  • The grade no longer means what it used to

“It becomes difficult to push back and say ‘you used AI’ — because we don't have a tool to check. I looked online, and all the tools I'm hearing are just not reliable.”

— Public Health Professor · Research University

Every keystroke tells a story.
Now you can read it.

UnProctor captures the full writing process — before AI, during, and after — so you have the context to make informed decisions.

📸

Version snapshots

See every meaningful draft, not just the final output. Know exactly what a student had written before they touched AI.

💬

AI interaction log

Know when and how AI scaffolding was used. Every prompt, every response — with flagging for answer-seeking behavior.

Submission comparison

Final answer vs. pre-AI draft, side by side. Added words highlighted. Words removed shown. Net change calculated.

🎚

Instructor-controlled constraints

You define what AI can and can't help with, per question. The assistant enforces your rules — not a generic policy.

Jordan Kim — Q2 Submission
Surveillance capitalism · ETH 210 Midterm
Submitted
C — Before / After Answer ComparisonQuestion 2 · Captured drafts

Before AI Access

Captured at: 14:22:10

Surveillance capitalism is an economic system in which corporations extract data from users not to improve services but to predict and shape future behaviour for sale to advertisers. The key distinction is between data collected to serve the user and behavioural surplus collected to serve advertisers.

After AI Access (Final)

Captured at: 14:52:30

Surveillance capitalism, as Zuboff describes it, is an economic system in which corporations extract behavioural data from users not merely to improve services but to predict and shape future behaviour for sale to third-party advertisers. The key distinction Zuboff draws is between data collected to serve the user and behavioural surplus collected to serve advertisers.

+47 added−0 removedNet: +47 words
In preview

Not a score. A signal.

UnProctor doesn't tell you if a student cheated. It gives you the information to have a real conversation.

Across your class roster, see a simple signal showing the ratio of original writing to post-AI revision. Surface outliers worth a second look — without generating false accusations.

Low signal — Student wrote most of the work independently, used AI to check reasoning.

Medium signal — Notable revision after AI use. Worth reviewing the before/after comparison.

High signal — Significant post-AI content added, plus flagged message. Open a conversation.

ETH 210 · Midterm Essay · 18 submitted · AI signal overview
AK
A. K.
22%
MC
M. C.
18%
JK
J. K.
58%
SR
S. R.
31%
TW
T. W.
84%
PL
P. L.
25%
RM
R. M.
47%
+ 11 more students

Built for instructors who care about the work, not just the output.

✍️

Writing-intensive courses

English, humanities, social sciences, law. Courses where process and original thought are the point — and where AI assistance most directly undermines the learning outcome.

🎓

Graduate programs

Where originality and intellectual development matter as much as the final product. Qualifying exams, dissertation proposals, seminar papers — the work that defines a scholar.

🏛️

Institutions navigating AI policy

For administrators who need a principled framework — not a blanket ban. UnProctor gives institutions the infrastructure to permit AI use responsibly, with full accountability.

Everything traditional proctoring got wrong.
And what we're doing instead.

Traditional proctoring tools
  • Webcam surveillance of students in their homes
  • AI detection scores with documented racial bias
  • False positive rates that punish innocent students
  • Adversarial framing — students as suspects
  • No insight into how students actually learn
  • Privacy concerns with third-party data sharing
UnProctor
  • Full writing process audit trail — no cameras
  • Process signals, not accusation scores
  • Designed to produce conversation, not verdicts
  • Faculty-first — instructors control every policy
  • Learning analytics, not surveillance analytics
  • Privacy by design — your data stays yours

Built with faculty, not for administrators.

6
Instructors interviewed across 5 disciplines
5
Fields represented: CS, HCI, Public Health, Earth Science, Project Management
100%
Reported difficulty detecting or proving AI use in submissions

I think where the AI has added complexity is — I am not sure if this is AI-written or not. If I can clearly say it’s AI-written, then it reduces the complexity. But I’m in this dilemma: I can’t point to students, because I don’t want to incorrectly blame them. I don’t want to go in that direction unless I have a way to confidently tell.

A
HCI Professor · CS Department · Research University
Onboarding first cohort now

Ready to see the work,
not just the output?

UnProctor is in early access. We're onboarding a small cohort of institutions for the first semester. No commitment required.

Institutional email addresses only · No spam, ever · Unsubscribe anytime

See it in action

Walk through a 3-step demo — from exam setup to student submission to the full audit trail.

View Demo