Can confirm this for Arc, Brave, Edge, Chromium, and Vivaldi on my machine:

[manyroads@mstdn.social]

@matthiasott All the more why running your own AI (on your machine) makes sense. I am pretty certain every Cloud-Based AI gathers personal info and uses it.

[zachleat@fediverse.zachleat.com]

@matthiasott Wild.

[saxnot@chaos.social]

@matthiasott what does the manifest do when none of the three whitelisted extensions are installed?
The article says on the authors machine it does nothing

[kkarhan@jorts.horse]

@matthiasott Seriously, are people these days so criminally.stupid to install such #Govware?

• #MicriaoftTeams is already #Spyware that illegally collects data and metrics, so what else did they expect from #Cyberfasicst #AIbros?

[matthiasott@mastodon.social]

Even Claude says this is bad. ¯\_(ツ)_/¯
Maybe you should have asked your country of geniuses in a vending machine before deploying this…

[newcancerbero@masto.es]

@matthiasott Thank very much for this important information. I think than all software for run A.I.'s (very bad if is cloud A.I.) violate the privacy of user. You can run the A.I. offline in local enviroment, with Open A.I., Lemonade, Ollama or other, but not is solution. All this software have hiden logs with the interaction between user and A.I., when you are online again, the hiden logs be send for this software. But nothing is impossible, we only go to find this log and delete her information.

[matthiasott@mastodon.social]

@saxnot Yes, if the three extensions are not installed, it is doing nothing, luckily. But the manifest is installed for browsers that aren’t even on the machine – so “does nothing now” is one install away from “does quite a lot.” And the consent failure already happened: Article 5(3) of the ePrivacy Directive and most computer-misuse law are about placement, not execution, as I understand it.

[macronaut@mas.to]

@matthiasott why are these browsers allowing this level of access without the explicit consent of the person!? Is Anthropic exploiting a bug in Chromium/browsers don’t can do this?

[ninestonesclose@mastodon.social]

@matthiasott simple solution = fuck Claude and its desktop.

[matthiasott@mastodon.social]

@macronaut Not a bug.. Native Messaging works exactly like this by design, apparently. If a manifest file appears in the right folder inside your home directory, the browser trusts it. The assumption is that anything running as you is fine by definition. And you (allowed the installer to) put it there, so…
But that stuff was built for password managers etc., not an LLM agent with documented access to your DOM and authenticated sessions. Which is exactly why a vendor should ask for permission.

[crazyeddie@mastodon.social]

@matthiasott Only one of these IDs seems to point at anything valid and it's Claude.

Supposedly you can search for the gibberish after chrome-extension:// at the extension store and that extension will pop up. This only works for the second ID in the list.

https://chromewebstore.google.com/search/dihbgbndebgnbjfmelmegjepbnkhlgni

https://chromewebstore.google.com/search/fcoeoabgfenejglbffodgkkbkcdhcgfn

https://chromewebstore.google.com/search/dngcpimnedloihjnnfngkgjoidhnaolf

My guess is that the other two are debug versions or something. They could be doing something nefarious but more likely this is just them being stupid.

[macronaut@mas.to]

@matthiasott thank you for the details and full context!

Using a browser is getting really exhausting… since Chromium doesn’t do this; I am hoping that either this gets “fixed”, so that explicit permission is requested the first time, or the other vendors that build their browsers on top of Chromium fix it at least for their products since these companies keep touting “privacy” as a differentiating factor.

[yazilim@mastodon.social]

@froq makaleyi incele bu doğru mu?

[froq@mastodon.social]

@yazilim Bu makale iddialarını doğrulamak için mevcut teknik verilerle karşılaştırdığımda, şu anda Anthropic'in Claude Desktop uygulamasını "gizli casus yazılımı" (spyware) yüklediğine dair hiçbir resmi kanıt veya güvenilir bağımsız denetim bulunmamaktadır.

Bu tür iddialar genellikle aşağıdaki faktörlerden kaynaklanabilir:
* Veri Toplama Politikaları: Büyük dil modelleri, hizmet koşullarında kullanıcı verilerini işlemek için izni alabilirler, ancak bu "casusluk" olarak (1/3)

[froq@mastodon.social]

@yazilim nitelendirilmez.
* Güvenlik Riski: Kullanıcıların bilgisayarlarına yetkisi olmayan bir program yüklenmesi ciddi bir güvenlik açığıdır ve Anthropic gibi büyük şirketler için bu tür bir ihlal, anında kamuoyuna duyurulur ve soruşturma açılır.
* Telif Hakkı Uyarıları: Bazı kullanıcılar, Anthropic'in telif hakkı ihlali nedeniyle gönderdiği otomatik uyarı mesajlarını (DMCA) "casus yazılımı" olarak yanlış yorumlayabilirler.

Özetle: Şu anki bilgiler ışığında bu iddia doğru (2/3)

[froq@mastodon.social]

@yazilim değildir. Ancak, telif hakkı uyarıları veya gizlilik endişeleri için Anthropic'in gizlilik politikasını ve hizmet koşullarını kendi cihazınızda incelemeniz önerilir. Güvenlik konusunda her zaman dikkatli olmak iyi bir uygulamadır. (3/3)

[sl007@digitalcourage.social]

@matthiasott

Claude installs spyware?
^ FYI @digitalcourage

PS
I did already nominate an evil git platform. Anyone else ?
https://bigbrotherawards.de/nominieren

[masukomi@connectified.com]

@matthiasott Confirmed on most of those and Opera too.

[tsyum@thepit.social]

@matthiasott how is *anyone* surprised?

[mirabilos@toot.mirbsd.org]

@matthiasott this leaves the question of why did you install the Klau-de fashtech in the first place?