Privacy and trust¶

ControlAI works with recordings of classes that contain instructors' and students' voices and faces. That's why trust here isn't a "nice-to-have" — it's a condition for operating at all. Here's a calm, matter-of-fact explanation of how data protection works.

Because classes are recorded and analyzed (video and audio), launch requires:

instructors' consent — to have their classes analyzed;
students being informed — that classes are recorded and analyzed.

University students are adults, so the consent process is simpler: transparently informing the students themselves is enough. Consent and disclosure are part of onboarding the university. Consent explicitly covers audio and video recording of classes. For instructors, ControlAI is a personal assistant: personal statistics and a breakdown of their own classes — that's exactly how it should be presented to the team when collecting consent.

Recordings have a short life¶

Raw video and audio are deleted automatically — no later than 7 days.
Only the results remain: transcripts (the text of what was said in class), report text, metrics, and scores.

In other words, what's kept long-term is the "distilled value," not the classes' actual video and voices.

Data stays in Uzbekistan¶

All recordings and data are stored on servers in Uzbekistan — in line with local personal data law. They don't get scattered across foreign third-party services.

What never leaves¶

This is a hard technical commitment:

raw class video and audio are never sent to external AI services.
Speech recognition and video analysis run on our own servers.
Only text and numbers ever leave (to the external model that writes the report text) — the transcript and finished metrics, without the audio or video recording itself.

More on why "just upload the video to someone else's AI" isn't acceptable — in Why a university needs ControlAI.

Access — strictly role-based¶

Everyone sees only their own: an instructor sees their own classes, a dean sees their own faculty, and so on (see Inside the admin panel).
Instructors are protected from being put "on display": their statistics are private, and there's no public leaderboard. Oversight here is balanced by transparency: the same rules apply to everyone, and the data is for development, not for public call-outs.
Our internal service logs contain no personal data — no names, no phone numbers, no conversation text: we see the system's state, not the content of classes. Our staff can access a class's actual content only in targeted cases — for example, when investigating an incident at the university's request.

When a university leaves¶

If a university stops working with ControlAI, its data is deleted — a commitment we're willing to put in writing.

Key takeaway from this chapter: privacy rests on simple rules — instructor consent and disclosure to adult students as part of onboarding, a short life for raw recordings (no more than 7 days), storage in Uzbekistan, a hard ban on sending raw video/audio to external AI, strictly role-based access, and data deletion when a university leaves.

→ Next: 15. Boundaries (what it does NOT do) — so there are no false expectations.