#photodna — Public Fediverse posts

Microsoft's PhotoDNA technology keeps flagging my face picture

https://www.elevenforum.com/t/microsoft-photodna-scanning-problem-it-is-comical-now.45961/

#HackerNews #Microsoft #PhotoDNA #facial #recognition #technology #scanning #issues #comical

Des chercheurs de la #KULeuven ont percé l’algorithme ultrasecret de #PhotoDNA, la technologie de #Microsoft utilisée par toutes les plateformes pour débusquer les pédocriminels. Quelques secondes suffisent pour le contourner. Ou accuser des innocents.

https://www.lesoir.be/735999/article/2026-03-21/le-systeme-mondial-de-detection-dimages-pedopornographiques-en-ligne-est-boiteux

https://eprint.iacr.org/2026/486.pdf

Let Me Explain How a State Actor Could Perform a Denial-of-Service Attack on the Entire UK Government in the Wake of Ofcom “Online Safety Act” Client-Side Scanning

1/ obtain a hash of abuse material that’s both known & banned; if pervasive as claimed this shouldn’t be hard

2/ use algorithms from this paper to create a cat meme with the same hash

3/ send the cat meme to all MPs & Civil Servants via SMS, E-Mail, WhatsApp (bonus if it goes viral)

4/ watch as MPs are locked out & banned by platforms for possessing abusive material, preventing government

Also: there is no mitigation by saying “all of these cases should be appealed” because by the time enough resources have been deployed to resolve the appeal claims, government will have been offline for 12 hours or more.

Of course one could propose mitigations: the government could ex-ante inform all platform providers which accounts needed to be prevented from ever being blocked – to be given special treatment – however:

1. Politically that will not fly well, and…
2. It will immediately cast doubt upon any politician who has an unacknowledged backup social media profile, which will be visible to the platforms of course because of device cookie sharing and things like that, so…
3. Such information will be a topic of extortion or leaks

Also, there will be claims of “two-tier surveillance” and so forth.

White-Box Attacks on PhotoDNA Perceptual Hash Function

https://eprint.iacr.org/2026/486

[*] note: strictly, it is not necessary to obtain the material, merely the hash; therefore a leak of the existing database of hashes – several million in size – would be catastrophic by providing material for an infinite sequence of attacks like this.

At this link, you can read the transcript, or listen to an audio recording, of an early August meeting attended by the Stanford operatives, a co-author of the ActivityPub standard, and a prominent fediverse developer, as they discuss their plans to impose centralized algorithmic surveillance onto the network.

https://github.com/swicg/meetings/tree/main/2023-08-04

Here's more on one of the surveillance systems under consideration, a technology controlled by Microsoft, which detects and auto-reports both CSAM and political subversion: https://kolektiva.social/@ophiocephalic/110782738969976772

4/20

#FreeFediverse #FediPact #DefederateMeta #Meta #Facebook #Threads #Prefiguration #DualPower #Fedifam #Communalism #FacebookFediverse #Zuckerverse #PhotoDNA #FacebookMafia

Followup to the PhotoDNA post: @dansup has reached out and confirmed that @pixelfed will NOT be enabling PhotoDNA. Thank you Daniel for your communication and your regard for privacy on the fediverse.

There are ideas circulating on decentralized, privacy-respecting ways to better keep everyone safe, ease burdens and liabilities for admins, and keep harmful material out of our timelines. These prospects would require more collective effort than simply signing up for a third-party service, but it's worth it. Our communities are overwhelmingly comprised of good people who aren't hurting anyone. We don't deserve to have algorithmic corporate surveillance inflicted on us.

Viva Pixelfed! :pixelfed:

https://mastodon.social/@dansup/110802145007075499

#FediPact #FediblockMeta #DefederateMeta #Meta #Facebook #Threads #FacebookFediverse #PhotoDNA #Surveillance #Pixelfed

Those considering the use of the PhotoDNA surveillance process in the fediverse are advised to inform themselves on it first. It doesn't just detect "positives", it can also auto-report them to authorities. Implementing it could turn this network into an automated police-state snitching system. Admins who do not choose auto-reporting may be legally obligated to manually report positives.

Also (independently verified and officially contradicted), PhotoDNA is utilized not just to detect CSAM, but also "terrorist and violent extremist content". So who decides what constitutes this content? Why, the private unaccountable black box which is Microsoft Corporation, that shining beacon of privacy-respect to which a hash of every image uploaded to and federated across enabled instances would be sent. Microsoft indicates the data can be utilized for facial recognition and "AI" ingestion as well.

Do they - or will they - rate advocacy of Palestinian equality and Kurdish feminism as extremist, as governments they do business with would demand? Do they or will they accept the inputs of oil and pipeline corporations in determining what constitutes terrorism? How will they expand the parameters of disallowed sexual deviance if Trump wins the election next year?

Understand that this is already happening. Elements are in the Mastodon Github Issues calling for the addition of PhotoDNA right now. The Pixelfed project may also be adding it very soon. If that is true, antiauthoritarian instances should consider the feasibility of defederation from Pixelfed altogether. [edit: Pixelfed is no longer considering the addition of PhotoDNA, confirmed in replies below.]

For all those considering this: Stop. Please. All the "report" issued by the Facebook-mafia describes is the fediverse - and in fact, mostly its evil twin, the defediverse - just as it existed last month, last year, years before most on here had ever heard of it. The purpose of their influence operation is scare everyone into turning the fediverse into the kind of policed and surveilled space suitable to Mark Zuckerberg.

Microsoft FAQ on PhotoDNA:
https://www.microsoft.com/en-us/PhotoDNA/FAQ

Microsoft Digital Safety Content Report, indicating use of PhotoDNA to detect extremism:
https://www.microsoft.com/en-us/corporate-responsibility/digital-safety-content-report

Technical analysis of the PhotoDNA process:
https://hackerfactor.com/blog/index.php?archives/931-PhotoDNA-and-Limitations.html

UN report referencing use of PhotoDNA to detect terrorism:
https://www.un.org/counterterrorism/sites/www.un.org.counterterrorism/files/countering-terrorism-online-with-ai-uncct-unicri-report-web.pdf

Transcript of 2018 US Senate hearing on terrorism and social media, referencing use of PhotoDNA:
https://www.govinfo.gov/content/pkg/CHRG-115shrg31316/html/CHRG-115shrg31316.htm

Journalism indicating use of PhotoDNA to detect extremism:
https://www.theatlantic.com/technology/archive/2016/06/a-tool-to-delete-beheading-videos-before-they-even-appear-online/488105/
https://www.scientificamerican.com/article/when-hatred-goes-viral-inside-social-medias-efforts-to-combat-terrorism/

Wikipedia pages:
PhotoDNA https://en.wikipedia.org/wiki/PhotoDNA
Global Internet Forum to Counter Terrorism https://en.wikipedia.org/wiki/Global_Internet_Forum_to_Counter_Terrorism

#FediPact #FediblockMeta #DefederateMeta #Meta #Facebook #Threads #FacebookFediverse #PhotoDNA #Surveillance

Stanford researchers find Mastodon has a massive child abuse material problem

Mastodon is rife with child sexual abuse material (#CSAM), according to a new study from Stanford’s Internet Observatory

In just two days, researchers found 112 instances of known CSAM across 325,000 posts on the platform — with the first instance showing up after just five minutes of searching.

Decentralized networks don’t use the same approach to #moderation as mainstream sites. Instead, each decentralized instance is given control over moderation, which can create #inconsistency across the Fediverse.

That’s why the researchers suggest that networks like Mastodon employ more robust #tools for moderators, along with #PhotoDNA integration and #CyberTipline reporting

https://www.theverge.com/2023/7/24/23806093/mastodon-csam-study-decentralized-network

A significant portion of the child abuse material researchers uncovered was from networks in #Japan, where there are “significantly more lax laws” that “exclude computer-generated content as well as manga and anime,” according to the report.

“We found that on one of the largest Mastodon instances in the Fediverse (based in Japan), 11 of the top 20 most commonly used #hashtags were related to #pedophilia,” the researchers wrote.

** For nuance read the report itself; https://stacks.stanford.edu/file/druid:vb515nd6874/20230724-fediverse-csam-report.pdf

Thanks @JDGooiker

I somehow missed that #FOSDEM had a keynote where #Facebook advertised its #uploadfilters which build on the infamous #PhotoDNA.

No mention of ethical considerations.

https://fosdem.org/2023/schedule/event/similarity_detection/

Q&A: can't confirm nor deny whether SSCD is used in production, whether Facebook does anything to correct false positives, whether it's possible at all (given it requires re-training).

#ChatControl #privacy #EUdataP