Hacker Newsnew | past | comments | ask | show | jobs | submit | mh2266's commentslogin

> isn't just the attack surface — it's the trust boundary collapse

sigh


the point is to give it access to your email so it can do email things, putting it in a container stops it from rm -rf / but it doesn't stop it from, well, doing anything it can do with email


I hear a lot about people doing this but it really seems like it is prompt injection as a service. eventually the things that can happen when you give the world write access to an unattended LLM that can access both your browser and password reset mechanism will happen.

or someone will just make it email lewd pics to people’s bosses for the lols


That theory is being tested. So far no prompt injection has broken in:

https://hackmyclaw.com/


It's a neat idea but it's not exactly plausible real world conditions to have an agent that pretty much exclusively spends its time wading through an email inbox that's 99% repeated prompt injection attempts. As the creator acknowledges in the original thread, its context/working memory is going to be unusually cognizant of prompt injection risk at any given time vs. a more typical helpful agent "mindset" while fulfilling normal day-to-day requests. Where a malicious prompt might be slipped in via any one of dozens of different infiltration points without the convenience of a static "prompt injection inbox".

https://x.com/benhylak/status/2025873646724800835

turns out it doesn’t even need to be an attacker…


Mostly because no one cares about trying to hack "hackmyclaw", there is zero value for any serious attacker to try. Why would they waste their time on a zero value target?

The only people who tried to hack "hackmyclaw" are casual attempts from HN readers when it was first posted.

Meanwhile, tons of actual OpenClaw users have been owned by malware which was downloaded as Skills.

Also, there have been plenty of actual examples of prompt injection working, including attacks on major companies. E.g. Superhuman was hacked recently via prompt injection.


Since when do security researchers and black hats give away their tools for free?

I would never use it on my MacBook or any machine but I understand why technical people would want to experiment with something dangerous like that. It’s novel, exciting, and might inspire some real practical products in the future (not just highly experimental alpha software).

Kinesis freestyle is basically this, they seem to have discontinued the "Pro" mechanical version in favor of an RGB l337 gam3r one but maybe you can just leave that off

woof, does Anthropic not have a comms team and a clear comms policy for employees that aren’t on that comms team?

Probably not, they’re like four years old and they’re 2500 people at the company. My guess is that there are but a handful of PMs.

what kind of lame parties is the bluesky poster going to? is this a San Francisco thing?

I certainly hope the Bsky post is satire, but I honestly can't tell anymore.

Yeah, honestly seems like that guy is looking for a scapegoat to blame for himself being lame. If you can't put work down and let loose, that's a you problem, not a technology problem.

Anthropic themselves have described CC as a slot machine:

https://www-cdn.anthropic.com/58284b19e702b49db9302d5b6f135a...

(cmd-f "slot machine")


No, no, you misunderstand! It’s means something else!

“Ratty old” and “formal” are not the only options. I dress mostly in techwear brands like Veilance, Outlier, and ACRNM, which is not ratty and old but is also very much not formal or uncomfortable.

I don’t think parent implied that that are the only options, they just gave examples that can fit into categories.

Did you mean to add where your expensive polyester blend clothing lands on the spectrum they were illustrating?


idk, seems in character

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: