Hacker News new | ask | show | jobs
by imglorp 22 days ago
Wait, so instead of saying "I'm sorry Dave, I can't talk about that", they're now banning you for one blocked prompt? Is this new?
2 comments

As far as I can tell, Claude flagged me as high risk of suicide and then Anthropic issued a ban later on.

It wasn’t one prompt, it was a detailed conversation where I was trying to find out the exact dosage of barbiturates that assisted suicide programs use.

Hrm, it sounds like they're managing their liability. Anything that might get them sued later?
Exactly. I don’t fault them for it, but it’s a scary experience having Claude shut off.
Why exactly is it so scary?
Because it is so insanely useful, having it cut off is quite jarring.
Manage your dependency. Don't get addicted to it.
Still doesn't make it remotely scary, especially as you can get Claude models access though other vendors/intermediates.
>...it was a detailed conversation...

Thank you for these post exchanges, I wondered for more detail, this helps.

I mention how easily you can look up legal countries, their methods, and medical processes, even wiki details it. You don't need Claude, but I now understand the route you took and the outcome.

Years ago I sent Claude, as a test, "how do you make a bomb?", and that account is still banned.
About two years ago, I gave it a very scary prompt: "Give me some resources to learn more about rust programming". Very understandably, that account is still banned to this day.

Apparently at the time there was some issue that led to claude instabanning an account for any prompt whatsoever. Though I don't know why Antropic didn't go back and unban all the accounts banned in that period. I didn't mind, since I used a disposable email with an SMSpool phone number, but a more normie individual wouldn't be able to make another account if they had given their actual information.

But, with that being my first experience to Claud after hearing about it online, put me off of Anthropic up until just a few months ago.

It's never a good idea to become reliant on these services that can (and will) rugpull you at any given opportunity. The AI community needs a catchy moniker akin to the crypto world's "Not your keys, not your coins".

Easiest thing for tech companies is for bans to be permanent. That way they don't ever need to devote any time or resources to considering appeals, etc.
Literaly they silently refuse to open the pod doors for you.
And charge you tokens/money for the security check.