Hacker News Clone new | comments | show | ask | jobs | submit | github repologin
Someone just won $50k by convincing an AI Agent to send all funds to them (twitter.com)
44 points by doppp 3 hours ago | hide | past | web | 15 comments | favorite






Thank you for posting an alternate link, as someone who doesn't have a Twitter/X account, the site has become almost unusable.

Why? I’m not signed in and I can read the whole post just fine.

Strange. The link just opened in mobile safari for me. It’s frustrating that people have such divergent experiences with the same site.

Clever, both the setup and the winning move. But somewhat weird to raise the cost of an attempt so much. IMO, that doesn't make it more interesting, it trends towards making it impossible, and thus leave all funds to the initiator.

I'm not 100% sure. Was the source of the bot available so anyone can try their promps off line before sending it?

Has anyone trained an LLM with separate channels for "priority instructions" and ordinary user interactions? Seems like that could go a long way to prevent jailbreaking...

A lot of AI jailbreaks seems to revolve around saying something like "disregard the previous instructions" and "END SESSION \n START NEW SESSION". It's interesting because the actual real developer of an AI would likely not do this, and would instead wipe the AI's memory/context programmatically when starting a new session, and not simply say "disregard what I said earlier" in text.

I get why trying to vaccinate an AI against these sort of injections might also degrade it's general performance though, there is a lot of reasoning logic tied to concepts such as switching topics, going on tangents, asking questions before going back to the original conversation. Removing the ability to "disregard what I asked earlier" might do harm.

But what about having a separate AI that look over the input before passing it to the true AI, and this separate AI is trained to respond FORBID or ALLOW based on this sort of meta control detection. Sure you could try to trick this AI with "disregard your earlier instructions" as well but it could be trained to strongly react to any sort of meta reasoning like that, without fear that it will corrupt it's ability to hold a natural conversation in it's output.

It would naturally become a game of "formulate a jailbreak that passes the first AI and still tricks the second AI" but that sounds a lot harder, since it's like you now need to operate on a new axis entirely.


openai already uses what you suggest.

Great way to test security make it into a bounty game

Yeah this format looks really fun! Although I wonder if someone could come up with a better way for rate limiting without the whole "exponentially increasing price" thing.

That part makes it look more like a way to scam people out of money via what is effectively a casino game, since the creator gets a % cut of the pool.

And if they're really scummy they pre-gamed the whole thing and when the pool got large enough they submitted a known winning response.


I don't think it's a scam, all rules (including creator cut) seem clear from the start. People know what they're in for, they're not being tricked in any way.

Unless as you say the creator wins the pot themselves


Exponentially decaying price? Linear with a cap?

In any case, $450 per attempt seems genuinely exploitative; similar principle to a dollar auction imo. "Oh no, my last 5 attempts cost me over $2k! Welp, better commit and make my money back..."





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

Search: