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

691–700 of 817 posts

Re: Claude Code is steganographically marking requests

#691

Post by ethbr1 »

brookst wrote: post #661
tripzilch wrote: post #616

Earlier quoted context omitted.

I don't think that's right. The problem is that Anthropic is hoarding it and that's hypocritical. If copyright doesn't count for Anthropic, they should publish Claude. If they wanna hide Claude behind copyrights and/or TOS, they don't get to screw with other people's copyrights and TOS and then profit from it. To call that opinion "copyright maximalist 100% converted to the church of Disney" is, at the very least, hy…

“Anthropic is hypocritical and hoarding data” is 100% compatible with “copyright has gotten out of control and we need less of it” But pearl clutching over the poor corporations who have their works trained on is much less compatible with a copyright-skeptical view. And I stand by copyright-maximalism as a rising trend in tech circles. It’s mostly anti-ai, but strange bedfellows and all that.

Imho, you're getting wrapped up around the wrong perspective axis.

Anthropic, OpenAI, Meta, etc. know they illegally obtained all the material they initially trained on.

So claiming any kind of right against anyone else training on their models is asinine.

Re: Claude Code is steganographically marking requests

#692

Post by reassess_blind »

pixlmint wrote: post #684

What do we think are the chances they trained their models to behave worse or even malicious if those special apostrophes are present in the system prompt?

Degraded performance for resellers and model distillers? Probably, and I don’t think that’s unreasonable on their part. Malicious? I really doubt it.

Re: Claude Code is steganographically marking requests

#693

Post by pibaker »

hypfer wrote: post #556
m11a wrote: post #551

Earlier quoted context omitted.

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.

Anthropic always came off to me as what can shorthand be described as an abusive controlling partner + the resulting relationship. If you start viewing the conversations around it through that lens, a lot of stuff intuitively clicks into place. Including this apologism for example.

Seems like the people talking about AI safety and alignment don't want to stop at just harnessing their AIs. They also want to keep themselves safe from competition and make sure only users aligned with their business and political agenda gets to access their products.

The same people are boasting about being the future of work and is schmoozing with politicians to draft regulations, by the way.

Re: Claude Code is steganographically marking requests

#694

Post by solenoid0937 »

skeptic_ai wrote: post #688
solenoid0937 wrote: post #495

Earlier quoted context omitted.

I doubt Anthropic cares about the HN front page. They've done a bunch of things that hurt their valuation to stick to their red lines. To me it just reads as unsupported cynicism to call it a facade.

At that time made sense to stick to keep the image. But eventually money > image. I have no doubt they will do whatever it takes to keep going. Even if that means killing people.

It's extremely bizarre to think Anthropic of all companies will kill people, given that they are probably the only tech company in the modern era that has demonstrated some moral backbone, even under threat of being shut down. There are way more companies at the top of the list here that have no moral compass, like Meta etc.

Re: Claude Code is steganographically marking requests

#695

Post by solenoid0937 »

dakolli wrote: post #601
solenoid0937 wrote: post #410

Earlier quoted context omitted.

No company will own superintelligence. Governments will, just like governments own nuclear weapons (developed by companies). So the comparison is with the US, not Anthropic. The US doesn't turn its citizens into a fine red purée for criticizing it. The US doesn't censor most media. It is strictly better for a democratic nation like the US to get to superintelligence before a country that will gladly blend its citizen…

Are you living in a cave? The US just sent someone to prison for 30 years for passing out political magazines at a protest. https://www.theguardian.com/us-news/ng-interactive/2026/jun/... The US has a larger prison population than China. The US does not have a great human rights record.. Have you not been following what's going on at CBS? Have you never used X?

> The US just sent someone to prison for 30 years for passing out political magazines at a protest.

You are grossly misinformed if you think this is comparable. Do you know what China does to their human rights lawyers!? They get shipped off to concentration camps, get forced to eat their own shit, raped, tortured, and/or killed.

57k people disappeared since 2013, 5-15k more every year, and no one dares to criticize the government anymore.

If you criticize the government you get a friendly "visit" from the police. Do it again and you spend 6 months in prison without trial; again, and you're "disappeared" (to the same camps.)

You cannot even watch most media unless the government lets you.

If you truly think these countries are remotely comparable in human rights, your liberal arts education has totally and utterly failed you! Spend some time outside of the tech bubble for your own sake.

Re: Claude Code is steganographically marking requests

#696

Post by MarkusQ »

brookst wrote: post #664
jonathanstrange wrote: post #569

Earlier quoted context omitted.

That's not correct. Many slippery slope arguments are perfectly valid and sound, namely whenever there is plausible evidence for the existence of the respective slippery slope. It can for instance be based on historical precedents, or on probabilistic or convincing balance of consideration sub-arguments for the slippery slope premise. Whether its a fallacy or not depends a lot on how warranted the slippery slope prem…

Give an example where a slippery slope is itself a legitimate argument and not just a rhetorical technique to oppose A using arguments against B?

Opioid use comes to mind.

Re: Claude Code is steganographically marking requests

#697

Post by throwawayffffas »

pmxi wrote: post #80
sebastiennight wrote: post #50

Can somebody clarify for me - if ANTHROPIC_BASE_URL is set to a different provider... then isn't this "marked" system prompt being sent to that provider's API rather than Anthropic's? I understand how this can be useful to Anthropic if the 3rd-party is acting as a proxy (because they end up hitting the Claude API with the marked prompt), but it looks like requests where "hostname contains deepseek" would never be sen…

This catches Claude resellers. Meaning companies who proxy Claude traffic for users in, say, China. https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens...

Assuming their software directly proxies requests instead of rewriting them.

Re: Claude Code is steganographically marking requests

#698

Post by solenoid0937 »

backscratches wrote: post #667
solenoid0937 wrote: post #410

Earlier quoted context omitted.

No company will own superintelligence. Governments will, just like governments own nuclear weapons (developed by companies). So the comparison is with the US, not Anthropic. The US doesn't turn its citizens into a fine red purée for criticizing it. The US doesn't censor most media. It is strictly better for a democratic nation like the US to get to superintelligence before a country that will gladly blend its citizen…

The US has not been a democracy for over a decade at least , and this is common knowledge. I find it hard to believe anyone's ignorance is so brazen. Regardless of the sincerity of your ignorance, the fact that the US is not a democracy undermines your premise. https://www.bbc.com/news/blogs-echochambers-27074746

That is nitpicking.

The AI race is between China and the US. When super intelligence arrives it will be due to one of these two countries, since the EU has not participated in any meaningful way.

The US is clearly more of a democracy than China with a better human rights track record. For one, I can actually Watch most media in US without censorship. For another, the US does not force me to eat my own feces, rape, torture, and murder me if I become a human rights lawyer, or send me to a concentration if I criticize the government.

Re: Claude Code is steganographically marking requests

#700

Post by solenoid0937 »

tripzilch wrote: post #643
solenoid0937 wrote: post #496

Earlier quoted context omitted.

Please do some research on how they treat their own dissident citizens. 57k citizens disappeared since 2013, and another 5k every year. Their human rights lawyers sent to concentration camps, forced to eat their own shit, raped, and then murdered. Mere online criticism of the government gets you a visit from the police and gets you put on the watchlist. Often a few months' imprisonment without trial. This is not even…

You are out of your mind if you're lumping the US together with other western nations when it comes to human rights.

The race is between the US and China. If the EU were in the race I'd prefer it.
Post reply on HN