# Practical Privacy and Anti-AI Techniques for Video Calls

> Source: <https://discuss.privacyguides.net/t/practical-privacy-and-anti-ai-techniques-for-video-calls/39966#post_1>
> Published: 2026-08-17 14:08:30+00:00

Been thinking about privacy during online video calls, especially with AI transcription scribes becoming more common in meeting platforms and confidential setting like doctors rooms. Curious to know what others have tried or heard about regarding this.

Some of the things I have been thinking about for Zoom/Teams meetings:

**Background wallpaper with a non-consent notice** - I mocked up the attached background wallpaper with a non-consent notice. Wondering if this has any practical weight, or if it’s mostly symbolic?
**Adversarial patterns in the background** - I’ve read about research into visual perturbations ([like Fawkes or Glaze](https://opendatascience.com/3-tools-to-safeguard-images-from-ai-scraping/) and Nightshade) that supposedly confuse AI classifiers. Not sure if these would survive video compression in tools like Zoom or Teams, or if the concept translates from static images to live video feeds at all. Would be interested to hear if anyone’s tested anything like this.
**Audio noise for transcription interference** - Came across some papers on adversarial audio ([VoiceBlock](https://interactiveaudiolab.github.io/project/voiceblock.html), [ASRJam](https://sites.google.com/view/impulse-response-asr-attack/home)) that claim to raise word error rates significantly. The whole idea of adding subtle noise to block AI scribes is intriguing, but I’m skeptical about whether it’d work in practice without ruining call quality. Anyone have experience with this?

I have so many more questions and interested to start a conversation. Here are some questions to get started.

- What legal standing a visible non-consent notice actually has under various privacy rules in jurisdictions like the EU, UK, US, Canada and Australia?
- Are adversarial visual/audio techniques worth pursuing, or do video codecs essentially neutralize them anyway?
- Are platforms actually blocking this kind of thing, or is there room to push back?
- What you’ve tested (successes or failures)?
- Tools that might help, or things to avoid?