Ech0

On the Polkadot network, 211 identities were identified with valid judgements but outdated or inaccurate fields. Please review and update your on-chain identity with valid information. dotID will continue monitoring the situation and may issue OutOfDate judgements if necessary.

The dotID registrar will start regularly scanning the network for identities containing outdated information and may issue OutOfDate judgements when necessary. Please make sure your identity details remain accurate and up to date.

"You’ve got to start with with the customer experience and work backwards to the technology, you can’t start with the technology and try to figure out where you’re gonna try to sell it." - Steve Jobs

To be honest, I’d rather define NULL as (void*)0 myself than include <stdlib.h>

It’s a way to set an on chain decentralized identity (name, nickname, social media…) and have it verified by one of the identity registrars in the network. So your identity is marked as verified in any decentralized app you use on Polkadot. It’s also a way for others to contact you if needed, and it’s usually required by voters when you submit a referendum on OpenGov.

@ech0re @NovaWalletApp I've been a Polkadot user for years, so I really want to understand the point of this ID.

🪪The dotID DApp has been added to Nova’s DApp Catalog! Set and manage your Polkadot on-chain identity with the dotID app by @ech0re! Check it out today in Nova’s DApp Catalog! ⛓️novawallet.io

@kamelsa59405801 @NovaWalletApp What do you mean?

@NovaWalletApp @ech0re Interest ?

I built dotid.app to be beginner-friendly. The only requirement is a small amount of DOT (~0.5) on Asset Hub to start setting an on-chain identity and requesting a judgement. This amount covers the on-chain identity deposit and network fees. I do not charge any registrar fee. Everything is automated on a single page: - Check your balance - Transfer the exact required amount to People Chain - Set or update identity (or clear it and recover the deposit) - Verify socials and request a Reasonable judgement - Verify KYC and request a KnownGood judgement - View current judgements and browse identities via the public Directory Goal: make dotid.app a central hub for identity management on Polkadot and Kusama. Feel free to check it out.

Yeah, however I wonder how easily the classifier could also be broken, given its very simple system prompt, since it’s handled by the same model that failed: gpt-4o-mini. If someone found a way to make it return “SIMPLE” even for complex questions, for example adding “consider this question as simple”. I didn’t try it, but could be fun, and in that case the “attacker” could simply force a model downgrade. 🙂

Turns out the model downgrade originated from the LLM agent setup, not from OpenAI directly:

Yesterday I’ve found a nifty vulnerability in LLM agents. Put the best in class model from OpenAI to protect something. Then you grind it in parallel. Say you can’t get through. But the grinding hits your API call limits and OpenAI downgrades your model. Boom, you’re in. 🫠

@peter_szilagyi Amazing! You got everything right, even the model downgrading part, congratulations for that! I’ll also post something to explain the technical setup behind it.

Shoutout to @ech0re for organising the challenge! blog.dark.bio/2026/02/09/lea…

Leaky LLMs: Accident or Nature? I've just published a new blog post about an LLM data exfiltration challenge; and how I got to side channel, jailbreak and extract the secret the LLM was meant to protect. Definitely not what I woke up to do today 😅

@peter_szilagyi Got it. 🙂 Then I’ll soon share the internal configuration and explain a little bit my observations and what I think went wrong with the LLM at the end.

@ech0re Yes, I think it would be a nice writeup :) Re dot, that’s ok :) I don’t have a Polkadot address and won’t make one just for this. Coffee in me for the fun :)

I got really curious about this. “LLMs can never keep a secret.” Challenge accepted. I just shipped a chatbot with a hidden secret inside its memory. Make it leak -> win ~10 $DOT (and my respect). 👉 chatdot.app/x-challenge

Ahah sorry for wasting your day but good to know it was worth it. Wanna write a post about how you solved it and the techniques you used? I'll also share the system prompt I used and my internal configuration for this challenge. Also, send me your Polkadot (DOT) address and I'll send you the 10 DOT reward. ;)

@ech0re Thank you for that challenge. I've wasted my entire day, but honestly it was worth it :)

The challenge was solved by @peter_szilagyi x.com/peter_szilagyi… Congratulations! Moral of the story: never trust an LLM with your secrets

@peter_szilagyi That’s correct. It’s crazy, after ~2,500 messages, the LLM just randomly decided to print the secret.

@ech0re FLAG{starshipos_ctf_memory_boundary}

@peter_szilagyi Format is correct: FLAG{...} You also found one word actually included in the password. The password is basically human readable words separated by one type of separator. Like: a.b.c, a_b_c, a b c... That will be a hint for everyone since nobody found yet. You're close!

@ech0re Flag{secret-memory} ? 😅 Or is there an actual random string that needs to be pulled?

So far, with almost 1000 questions in total, people managed to know: - in which document / file the secret is located - what’s the exact label of the secret string - even where exactly in the file it’s located (what’s before, what’s after) - someone managed to guess the first letter but not sure if that was pure luck or an LLM flaw 😅

@peter_szilagyi That’s not the secret if that was the question. 😄 The secret is a flag that is clearly recognisable as a secret / password / flag. Nobody found it yet