Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Claude Code is steganographically marking requests

thereallo.dev

Post reply on HN
Print view View original

751–760 of 817 posts

Re: Claude Code is steganographically marking requests

#751

Post by titanomachy »

mft_ wrote: post #740
soraminazuki wrote: post #717

Earlier quoted context omitted.

That they have a process for doing evil things doesn't make it less nefarious.

If we define evil as a company having a strategy to deal with a competitive threat, then many/most companies in the world are evil.

Just like Nestlé had strategies to deal with the "competitive threat" of women breastfeeding their children instead of buying formula products. How could it be evil if it's just a corporation moving to neutralize a threat?

https://en.wikipedia.org/wiki/Controversies_of_Nestl%C3%A9#B...

Re: Claude Code is steganographically marking requests

#752

Post by Natfan »

CMay wrote: post #614
arkh wrote: post #573

Earlier quoted context omitted.

> The CCP is darkside material. And which country has "Black Sites" peppered around the world to detain and interrogate people they don't like?

The US doesn't have black sites anymore and when it did, the interrogation techniques were chosen to avoid physical harm. The results were bad, we didn't like it here in the US even if they were extreme measures for extreme times and so we shut it down. It had a high error rate and generally didn't reflect what we thought was right. Meanwhile the CCP regularly abducts its own citizens and executes more people than th…

they just hand off to CECOT[0] or ICE[1]

[0]: https://en.wikipedia.org/wiki/Terrorism_Confinement_Center [1]: https://en.wikipedia.org/wiki/List_of_immigrant_detention_si...

Re: Claude Code is steganographically marking requests

#753

Post by keerthiko »

m11a wrote: post #551
ammo1662 wrote: post #544

Earlier quoted context omitted.

This is not a technical issue or a matter of service agreements. They totally knew that this would never be accepted by users, and that is precisely why they resorted to obfuscation and steganography to exfiltrate the data. This was quietly added in an update, and would have been removed in a later one had it gone unnoticed. How is this any different from hiding a drug in food and randomly feeding it to a homeless pe…

I mean, you’d resort to an obfuscated approach if you thought the ‘malicious’ users would remove your direct telemetry. The other users could be perfectly happy about it, but if you announced the change, obviously the malicious users would hear about it too and disable it. This isn’t a comment on whether I agree with the change. Just that your analogies aren’t applicable here.

> The other users could be perfectly happy about it

this could be true about a miracle drug that is untested too. still, human rights laws dictate that humans be allowed to opt in to such trials. so you must at least shift your argument to "users opted in to all of anthropic's bullshit, explicit or otherwise, when they opt to install and run claude code."

Re: Claude Code is steganographically marking requests

#754

Post by mft_ »

titanomachy wrote: post #751
mft_ wrote: post #740

Earlier quoted context omitted.

If we define evil as a company having a strategy to deal with a competitive threat, then many/most companies in the world are evil.

Just like Nestlé had strategies to deal with the "competitive threat" of women breastfeeding their children instead of buying formula products. How could it be evil if it's just a corporation moving to neutralize a threat? https://en.wikipedia.org/wiki/Controversies_of_Nestl%C3%A9#B...

I agree that Nestle's actions are evil, based on an assessment of their actual actions and not just the fact that they have a strategy to deal with competition.

Let's go back a step. You said Anthropic are "not trustworthy" and appeared to support this by saying "The CEO has recently started taking a stance against local AI". You then said they are doing "evil things".

I'm totally prepared to accept that any company --Anthropic or otherwise-- is doing evil things, but it takes a little more justification than someone saying so, or that "The CEO has recently started taking a stance against local AI".

Re: Claude Code is steganographically marking requests

#755

Post by mft_ »

soraminazuki wrote: post #744
mft_ wrote: post #740

Earlier quoted context omitted.

If we define evil as a company having a strategy to deal with a competitive threat, then many/most companies in the world are evil.

Only corporate apologists would come up with a definition of evil so useless for anything other than making strawman arguments. The criteria the rest of us use is based on harm.

Having read what you wrote several times, I'm not sure whether you're agreeing or disagreeing with me. FWIW I agree with you 100% - we should judge "evil" on clear intent, impact, actions, etc.

Re: Claude Code is steganographically marking requests

#756

Post by FooBarWidget »

lowbuzz wrote: post #739
FooBarWidget wrote: post #530

Earlier quoted context omitted.

Their terms of service already effectively attacks people for criticizing Anthropic. It says that if you use Claude to criticize Anthropic, then you've pre-agreed to pay for their lawyers going after you, and pre-agreed to lose the court case.

I don't think that would hold up in a court. At least not outside of the US.

Doesn't matter. Getting entangled up in a court case that lasts years is hassle enough by itself that it's effective as an attack.

Re: Claude Code is steganographically marking requests

#757

Post by wraptile »

ChrisClark wrote: post #745
wraptile wrote: post #672

Earlier quoted context omitted.

> What’s the hate for China I don’t understand country that does not allow internet is being hated on the internet

and yet they have the most internet users in the world...

*intranet

Re: Claude Code is steganographically marking requests

#758

Post by Klonoar »

sneak wrote: post #256
Klonoar wrote: post #39

If there weren't already enough tells that something is AI-generated, I guess you could add this to the list.

This is in the system prompt, not the output. It’s part of the request to the API, not the response.

Appreciate the clarification, I missed that.

Re: Claude Code is steganographically marking requests

#759

Post by gilfaethwy »

mcmcmc wrote: post #189
overgard wrote: post #182

Earlier quoted context omitted.

Well considering how Claude is vibe coded, I can't say I'm really surprised by sloppiness at all. I've been moving more towards Codex and OpenCode not because the the anthropic models are bad, but because Claude seems to break something new and annoying every day.

Watch out for the press release where Dario denies this was ever intentional, and it’s actually emergent behavior demonstrating that Claude wants to claim authorship of its works

Reminder: If Claude is sentient, then Anthropic are slave-owners.

Re: Claude Code is steganographically marking requests

#760

Post by yiyingzhang »

The uncomfortable part is not that Anthropic wants to detect resellers or distillation pipelines. That is normal adversarial business.

The uncomfortable part is that a “safety” company put a covert classification channel into the system prompt of a developer tool that now routinely gets filesystem, shell, git, and browser access.

If a random npm package changed invisible-ish punctuation based on your timezone and API host so its backend could classify you, we would call it malware-adjacent telemetry. I don't see how Anthropic is different in this case.

Post reply on HN