# Privacy tools against facial recognition

**URL:** https://discuss.privacyguides.net/t/privacy-tools-against-facial-recognition/7242
**Category:** Questions
**Created:** 2022-10-28T09:21:15Z
**Posts:** 4

## Post 1 by @woheba — 2022-10-28T09:21:15Z

What are your thoughts on software that tries to counter facial recognition AI (like Fawkes ([GitHub - Shawn-Shan/fawkes: Fawkes, privacy preserving tool against facial recognition systems. More info at https://sandlab.cs.uchicago.edu/fawkes](https://github.com/Shawn-Shan/fawkes))) ? Is it efficient ? What are the best tools available ?

---

## Post 2 by @anon30510143 — 2022-10-28T12:23:35Z

The only problem I see with something like this is that it will always be a cat and mouse game. When you make a tool to defeat facial recognition, the facial recognition will just improve to counter it. I think the best way to avoid facial recognition picking up your face from pictures is to just block out the face in the picture.

---

## Post 3 by @woheba — 2022-10-28T12:37:00Z

Thanks for the reply ! That might work in certain situations but some of us have to upload a profile picture in LinkedIn for example, so it doesn’t make sense to block out the face…

---

## Post 4 by @anon34719932 — 2022-10-28T18:08:21Z

I think [threat modeling](https://www.privacyguides.org/basics/threat-modeling/) is the most important thing in these kinds of cases. If you decide to or are **required** to make certain information publicly available, then it’s helpful to think about the likelihood of unacceptable consequences.

In the case of AI deep learning, this seems like an interesting project and I agree with @anon30510143 concerning the cat and mouse issue. To my knowledge, we haven’t looked at tools like Fawkes enough to make a reliable evaluation of them yet, but the research is really interesting and I’m going to look for secondary sources on its approach.

In the meantime, if the minor quality degradation is acceptable to you, the pixel-level alterations the authors describe don’t seem to have much of a downside. You should ensure that you trust the tool, but—even if it’s ineffective—I can’t see any immediate issues.

Like I said, I’ve only taken a glance, so this **isn’t** an endorsement.

* * *

Update: I’m starting with this paper, so thanks @woheba for bringing the field to my reading list! :slight_smile:

[https://www.semanticscholar.org/paper/On-the-(Im)Practicality-of-Adversarial-Perturbation-Rajabi-Bobba/99ef1c7f42cdad2ee5474adc615edbcd495fc2f7](https://www.semanticscholar.org/paper/On-the-(Im)Practicality-of-Adversarial-Perturbation-Rajabi-Bobba/99ef1c7f42cdad2ee5474adc615edbcd495fc2f7)
