Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

I was banned from Claude for scaffolding a Claude.md file?

hugodaniel.com

Post reply on HN
Print view View original

131–140 of 664 posts

Re: I was banned from Claude for scaffolding a Claude.md file?

#131

Post by pixl97 »

swiftcoder wrote: post #49
eightysixfour wrote: post #45

Earlier quoted context omitted.

> Out of all of the different things these agents can do, surely most forms of "routine" customer support are the lowest hanging fruit? I come from a world where customer support is a significant expense for operations and everyone was SO excited to implement AI for this. It doesn't work particularly well and shows a profound gap between what people think working in customer service is like and how fucking hard it ac…

> shows a profound gap between what people think working in customer service is like and how fucking hard it actually is Nicely fitting the pattern where everyone who is bullish on AI seems to think that everyone else's specialty is ripe for AI takeover (but not my specialty! my field is special/unique!)

As someone who does support I think the end result looks a lot different.

AI, for a lot of support questions works quite well and does solve lots of problems in almost every field that needs support. The issue is this commonly removes the roadblocks from your users being cautious to doing something incredibly stupid that needs support to understand what they hell they've actually done. Kind of a Jeavons Paradox of support resources.

AI/LLMs also seem to be very good at pulling out information on trends in support and what needs to be sent for devs to work on. There are practical tests you can perform on datasets to see if it would be effective for your workloads.

The company I work at did an experiment on looking at past tickets in a quarterly range and predicting which issues would generate the most tickets in the next quarter and which issues should be addressed. In testing the AI did as well or better than the predictions we had made that the time and called out a number of things we deemed less important that had large impacts in the future.

Re: I was banned from Claude for scaffolding a Claude.md file?

#132

Post by szundi »

omer_balyali wrote: post #105

Similar thing happened to me back in November 19 shortly after GitHub outage (which sent CC into repeated requests and time outs to GitHub) while beta testing Claude Code Web. Banned and appeal declined without any real explanation to what happened, other than saying "violation of ToS" which can be basically anything, except there was really nothing to trigger that, other than using their most of the free credits the…

[dead]

Re: I was banned from Claude for scaffolding a Claude.md file?

#134

Post by vimda »

cortesoft wrote: post #5

I am really confused as to what happened here. The use of ‘disabled organization’ to refer to the author made it extra confusing. I think I kind of have an idea what the author was doing, but not really.

Yeah, referring to yourself once as a "disabled organisation" is a good bit, referencing anthropics silly terminology. Keeping it for the duration made this a very hard follow

Re: I was banned from Claude for scaffolding a Claude.md file?

#135

Post by hotpotat »

landryraccoon wrote: post #38

This blog post feels really fishy to me. It's quite light on specifics. It should have been straightforward for the author to excerpt some of the prompts he was submitting, to show how innocent they are. For all I know, the author was asking Claude for instructions on extremely sketchy activity. We only have his word that he was being honest and innocent.

I understand where you’re coming from, but anecdotally the same thing happened to me except I have less clarity on why and no refund. I got an email back saying my appeal was rejected with no recourse. I was paying for max and using it for multiple projects, no other thing stands out to me as a cause for getting blocked. Guess you’ll have to take my word for it to, it’s hard to prove the non-existence of definitely-problematic prompts.

Re: I was banned from Claude for scaffolding a Claude.md file?

#138

Post by ankit219 »

kstenerud wrote: post #110
ankit219 wrote: post #91

Earlier quoted context omitted.

My rudimentary guess is this. When you write in all caps, it triggers sort of a alert at Anthropic, especially as an attempt to hijack system prompt. When one claude was writing to other, it resorted to all caps, which triggered the alert, and then the context was instructing the model to do something (which likely would be similar to a prompt injection attack) and that triggered the ban. not just caps part, but that…

> When you write in all caps, it triggers sort of a alert at Anthropic I find this confusing. Why would writing in all caps trigger an alert? What danger does caps incur? Does writing in caps make a prompt injection more likely to succeed?

from what i know, it used to be that if you want to assertively instruct, you used all caps. I don't know if it succeeds today. I still see prompts where certain words are capitalized to ensure model pays attention. What i mean was not just capitalization, but a combination of both capitalization and changing the behavior of the model for trying to get it to do something.

if you were to design a system to prevent prompt injections and one of surefire ways is to repeatedly give instructions in caps, you would have systems dealing with it. And with instructions to change behavior, it cascades.

Re: I was banned from Claude for scaffolding a Claude.md file?

#139

Post by tomwphillips »

The post is light on details. I'd guess the author ended up hammering the API and they decided it was abuse.

I expect more reports like this. LLM providers are already selling tokens at a loss. If everyone starts to use tmux or orchestrate multiple agents then their loss on each plan is going to get much larger.

Post reply on HN