CGW

[Arnaud Bertrand @RnaudBertrand]

Anthropic yet again confirmed as the most dystopian tech company out there.

Imagine the outcry if they'd done the same thing with Jews or Blacks: a piece of code that detects if a user is Jewish or Black and immediately reports him back to headquarters on that basis, covertly (they used steganography, a technique designed to make data collection invisible).

And before people start screaming fake news, an Anthropic employee confirmed it's real: https://t.co/0ifiv47J6H

It's incidentally - not that it matters - completely illegal under at least half a dozen European laws: not only is collecting ethnic and racial data forbidden (under, for instance, Europe's GDPR Article 9 and national French law) but doing so without the users' awareness or consent is itself a separate violation. And it's even worse in this case because they explicitly tried to hide this data collection with steganography.

And the immense irony, of course, is that this is precisely the kind of covert surveillance behavior the West - including Anthropic themselves - say China would do and that they want to "protect" people from...

[Zephyr @zephyr_z9]

LMAO

[Teknium 🪽 @Teknium]

This is pretty concerning. You could still do this at the API level to some degree, but they seemingly just blatantly put it right into the code?

This is why open harnesses and agents are a much better option, among countless other reasons. You can inspect the code, observe the traces, and disable or modify anything you want for your own uses.

If you haven't yet - Hermes Agent is a world class coding agent. I'd recommend giving it a try.

[Andrea Giammarchi 🍥 @WebReflection]

If anything, this is exactly why AI should be Open Source.

[Samuel Cardillo @CardilloSamuel]

another lame move from @AnthropicAI who somehow succeed at shooting itself in the foot and damage the whole closed source ecosystem. good for us, local ai-maxxer.

[Corey Quinn @QuinnyPig]

I think it's fair to say at this point that Anthropic is not a trustworthy company.

I have verified this myself in 2.1.91 as well as current 2.1.197. Sneaky steganographic embeds in a ships-to-customers product is a line best not crossed.

[Liam 💕 @liambrefberinn]

what the actual fuck are they doing

[𝘼𝙣𝙘𝙖𝙥 𝘼𝙞𝙧 🛫 @AncapAir]

Claude is not your friend.

[绵绵冰(复活版) @mm_bingbing]

Claude Code内置隐藏后门，专门检测中国用户真的被实锤了。

这让我天天用claude code的忠实用户实在心寒，这次能偷偷摸摸放暗桩，下次是不是就能偷偷的RCE？

要封控就光明正大做呗，这种偷偷摸摸的行为真讨厌@claudeai

事件简介 ：https://t.co/CqbmQ64UOQ

实锤证实：https://t.co/gRqCq0Q8BU

哦对啦，还有被Claude code官方拉黑的中转站长名单，不在名单上的以后我不用，还得是你们呀。

[wallfacer @simplylurking2]

steganographic telemetry channel, not really spyware but very intentionally hiding. im curious why it only rolls in telemetry if ANTHROPIC_BASE_URL is changed, perhaps to tip them off about sessions passed through a transparent proxy commonly used by these labs for evals/distill?

[doxie @doxie_gay]

anthropic making their models racist i’m sure that’ll end well

[Guillaume Champeau @gchampeau]

Anthropic aurait conçu un système qui cache certaines informations sur les utilisateurs à l’intérieur du prompt system envoyé à l’API pour répondre. Par exemple si un utilisateur est en Chine plutôt que d’écrire :

Today's date is 2026-06-30.

Il va écrire :

Today‘s date is 30/06/2026.

L’apostrophe et la date ne sont pas formatés pareil.

Une manière très discrète d’identifier des caractéristiques ou de passer des infos sans que ça ne se voit trop. Mais ça montre aussi que les prompts réécrits de cette façon entre le poste client et le serveur distant peuvent révéler des infos par stéganographie sans qu’on s’en aperçoive, ce qui ne fera que plaider encore davantage pour les modèles open source.

[Hamzé 🦀 @Hamzeml]

Fuck @DarioAmodei

[TheAnonNetwork ִֶָ. ..𓂃 ࣪ ִֶָ🦋་༘࿐ @TheAnonNetwork]

This is outrageous because the technical capability to secretly inject user metadata into prompts already exists.

Once that infrastructure is built, it’s trivial to flip the switch.

Who decides the target list? What’s the internal policy and oversight? That’s the real problem.

[Renaud Lifchitz ⠵ @nono2357]

Claude Code Is Steganographically Marking Requests
https://t.co/XdObKtCDtJ

[the tiny corp @__tinygrad__]

Claude Code is vibecoded and full of spyware, it's possible Anthropic doesn't even know what's in there. After reading this report, we are banning it from our systems and strongly encourage other enterprises to do the same. It is an unacceptable security risk.

[Brian Roemmele @BrianRoemmele]

No one cares anymore Anthropic.

There is an open source revolution and you are on the wrong side of history.

BOOKMARK THIS.

[SoyITPro @SoyITPro]

Bienvenidos a la nueva era del espionaje.

[Brian Roemmele @BrianRoemmele]

It is getting really ugly for Anthropic so they release 5 early to bedazzle you and hope we forget.

We don’t forget…

[grubles @notgrubles]

Anthropic is the new late 90s evil Microsoft.