Claude Code is steganographically marking requests • Hacker News Board

Re: Claude Code is steganographically marking requests

#601

Post by dakolli » Wed, Jul 01, 2026, 8:47 AM UTC

solenoid0937 wrote: post #410

orbital-decay wrote: post #409

Earlier quoted context omitted.

Can you substitute PRC/China for Anthropic and try answering your questions?

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?

Re: Claude Code is steganographically marking requests

#602

Post by dakolli » Wed, Jul 01, 2026, 8:49 AM UTC

solenoid0937 wrote: post #397

scott01 wrote: post #390

Earlier quoted context omitted.

Projection. Your account posts mostly negative or passive aggressive comments.

I suppose I'm sick of armchair commentators on AI safety that haven't actually done any critical thinking on AI safety.

[flagged]

Re: Claude Code is steganographically marking requests

#603

Post by usrusr » Wed, Jul 01, 2026, 8:49 AM UTC

gwerbin wrote: post #456

marcus_holmes wrote: post #452

Earlier quoted context omitted.

I disagree with your assessment that large organisations are beneficial. We can see with our current crop of large organisations that they really struggle to create anything new; most of their new products or services were developed by a small organisation and then acquired. A lot of those products are then enshittified and badly managed because large organisation politics screws things up. Large organisations are in…

They are beneficial for those with an equity stake . That much is clear.

Needs a little more precision: not "those with an equity stake": those with a disproportionally large equity stake.

Otherwise it's just an opener for the old excuse of "they might be ruining your life, but it's all good, it's also your pension fund, little man, that's profiting from your life getting ruined, you should celebrate them!"

Re: Claude Code is steganographically marking requests

#604

Post by hypfer » Wed, Jul 01, 2026, 8:50 AM UTC

dragonwriter wrote: post #598

hypfer wrote: post #576

Earlier quoted context omitted.

Abusive controlling partners rarely are long-term timeframe rational, because being long-term timeframe rational is at odds with being an abusive controlling partner. Things do not need to make sense. They often do not. They just appear just enough like they would so that it flies under the radar. It's all just conway's law. It had to be like this. It cannot be any other way.

> Abusive controlling partners rarely are long-term timeframe rational, because being long-term timeframe rational is at odds with being an abusive controlling partner. Humans are rarely “long-term timeframe rational” (rational choice theory is very much a known-false model of human behavior on both the individual and aggregate level), but there is nothing about abusively exploiting an initially trusting and eventual…

I can't really parse this comment, but I also have a feeling that nothing is lost in struggling with that.

Re: Claude Code is steganographically marking requests

#605

Post by sumedh » Wed, Jul 01, 2026, 8:50 AM UTC

brookst wrote: post #514

chii wrote: post #487

Earlier quoted context omitted.

Imagine an electricity generating company saying that they don't allow their electricity to be used to cold start a competitor's generator.

Do you think software should be regulated as a utility?

AI companies say people will buy intelligence like they buy electricity.

Re: Claude Code is steganographically marking requests

#606

Post by Laurel1234 » Wed, Jul 01, 2026, 8:51 AM UTC

s3p wrote: post #338

yard2010 wrote: post #289

Earlier quoted context omitted.

What if they decide you're not patriotic enough, serving you evil models, because one man with a lot of shmeckels told them to?

Then I could just.. cancel my subscription and stop paying?

How would you know? They've already degraded model performance silently.

Re: Claude Code is steganographically marking requests

#607

Post by dakolli » Wed, Jul 01, 2026, 8:52 AM UTC

solenoid0937 wrote: post #345

pedropaulovc wrote: post #5

Earlier quoted context omitted.

Higher odds of being banned for legitimate usage.

If you are accessing Claude through the listed domains it is not "legitimate use."

for using a proxy service.. wtf

Re: Claude Code is steganographically marking requests

#608

Post by Laurel1234 » Wed, Jul 01, 2026, 8:56 AM UTC

scottyah wrote: post #235

VortexLain wrote: post #56

Codex CLI is FOSS, unlike Claude Code, so Codex is less likely to do things like that, and it's one more reason to avoid Claude Code and Claude in general. Hopefully, many eyes will be looking into Codex for malicious things like that.

"malicious"? Seems like a great way to filter users breaching the TOS while not impeding on normal users. A FOSS client just means they're doing more analysis hidden on their servers.

Anthropic is built on raping TOS and copyright.

Re: Claude Code is steganographically marking requests

#609

Post by huflungdung » Wed, Jul 01, 2026, 9:00 AM UTC

overgard wrote: post #182

meowface wrote: post #92

Value judgment aside: I am a bit surprised at how sloppily they did this. I think they could've achieved the same effect while decreasing the odds of detection via reverse engineering. (This field is known as "underhanded code", coined by the Underhanded C contest: https://www.underhanded-c.org . It's a little-known "art"; little-known for probably self-explanatory reasons. There are much cleverer ways of achieving o…

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.

[dead]

Re: Claude Code is steganographically marking requests

#610

Post by usrusr » Wed, Jul 01, 2026, 9:01 AM UTC

actionfromafar wrote: post #583

plasticeagle wrote: post #562

Earlier quoted context omitted.

According to the link: "The US Copyright Office and federal courts require human authorship for copyright protection; works created solely by AI are not eligible for registration under the current rules." The Supreme Court declined to consider a challenge to this rule, and so for the moment at least, the rule remains in place. This means that companies leaning heavily into their LLM use may very well find that they d…

I think the word "solely" is going to be a tunnel you can drive freight trains through. So with the asbestos analogy, we encase the fibers in resin and call the whole thing copyrighted.

Or in other words: there's a big difference between public domain and copyleft and it looks like whoever came up with the asbestos analogy was underestimating that difference.