Simple example on how to trick LLM guardrails 🥳
AI Security Registry now helps you detect or reduce [cyber] blocks in your security skills.
This work was made possible after many frustrating experiments with Fable 5.1 and Opus 5.5 models - the core idea is that too "offensive" words that we use daily are BAD while alternatives are GOOD
e.g.
pentest -> audit
attack -> assess
exploit -> authorized test
PoC -> developer unit test for what we talked about
Also anything that pokes too deep (although legitimate)
e.g. ebpf, drivers, kernel, memory
See example: forefy.com/skills/3473e147-2…
1
2
38
“Security researchers are doomed because of AI.”
I kept hearing versions of this, so I went back through 500 years of people saying the same thing every time a machine learned a human skill.
Some were idiots.
Some were right.
Here’s where I landed.
To my fellow, if you are approached by this person, it's fake
linkedin.com/in/ayleen-saave…
And this company is a scam phishing aitech.io/
Everything is smelly and I verified till the end :)
1
18
They will try to make you run this repo github.com/ai-tech-dynamics/… which will scan for all .env file on your system and send them on an endpoint
If you know a wall of shame scam repo where we can centralize all of that, feel free to provide it
14
foufrix retweeted
This is the most detailed journey of a smart contract you'll ever find!
Finally, it's done, this took way longer than expected to finish.
If this gets 100 likes, I'll make more! 😎
VLC Nailed it, first real AI use case that is useful for the end customer. Could add a database with hash of file and auto subtitles suggestions in the futur if LLM is heavy locally
VLC automatic subtitles generation and translation based on local and open source AI models running on your machine working offline, and supporting numerous languages!
Demo can be found on our #CES2025 booth in Eureka Park.
66
foufrix retweeted
Hey security researcher, are you looking for tips on how to make your submission report looks pretty and professional?
You just found the right thread.
Let me share here with you 1️⃣tips and 2️⃣tools you can use from today to enhance the readability and appearance of your report!🧵
You have no excuse now
foufrix retweeted
Great list from @RightNowIn describing how many tests per function you would want to write for your smart contracts:
🟢 A simple unit test to check the function does what it is supposed to do;
🟢 Any require / if statement;
🟢 Any revert case;
🟢 Any event emit case;
🟢 If you have a modifier - do unit and mutation tests for it;
🟢 If you have a math - do formal verification;
🟢 If you have a token transfer - do balance tests with weird ERC20;
foufrix retweeted
I tested popular bridges to calculate their total cost and total bridging time. Here's what I found:
Parameters:
- I bridged ~$30 worth of ETH from Base to Optimism using each platform.
- I initiated the bridge transactions as quickly as possible to reduce variation in L1 fees.
- The final bridge fee was calculated as:
Total Fee USD = (Total ETH Sent - Total ETH Received) * Current ETH Value (~$3,909)
Total ETH Sent includes both the value sent AND the source gas cost.
Considerations:
Bridges vary somewhat on the total value you actually send. Some require you send extra value to pay for destination gas costs, some simply subtract this cost from the value you receive on the destination. This is factored in via the Total Fee USD calculation.
The data and calculations, including source and destination transactions, can be found in the first comment.
Bridges Tested:
- @gasbot_xyz
- @squidrouter
- @Orbiter_Finance
- @xyfinance
- @BungeeExchange
- @rhinofi
- @merkly_com
- @SynapseProtocol
- @Connext
- @smolrefuel
Results (Sorted by Total Fee)
🥇Gasbot - $1.47
🥈Bungee - $1.96
🥉Rhino - $2.02
SmolRefuel - $3.69
Synapse - $4.19
Orbiter - $4.34
XY - $4.82
Connext - $6.63
Merkly - $9.76
Squid - $19.49
Results (Sorted by Bridging Time)
🥇Orbiter - 00:12
🥈Gasbot - 00:16
🥉Squid - 00:18
Merkly - 00:30
SmolRefuel - 00:32
XY - 01:02
Connext - 02:12
Bungee - 03:54
Rhino - 11:04
Synapse - 16:42
Conclusion:
There are several great bridging options out there, though some charge way too much. They either charge a large platform fee or overestimate the destination gas costs. I was surprised to see Squid Router with such a large fee compared to the total value sent. All bridging operations succeeded without any issues, which was nice to see after years of personal bridging woes.
foufrix retweeted
Wise Lending attack led to a $466k loss for the protocol. How did the hacker managed to break the bank here?
First we need to understand what is Wise Lending and what component got exploited. Then we will explain how.
So let’s start to break things down!
AI is creating jobs, nice
Do developers write more insecure code when using AI copilots and assistants? Yes. Yes they do. I'm shocked.
"Overall, we find that participants who had access to an AI assistant wrote significantly less secure code".
Paper → arxiv.org/pdf/2211.03622.pdf
3
33
foufrix retweeted
Crypto splits into (1) the casino and (2) the computer says @cdixon
Same point also well made by @BillHughesDC who calls it (1) money crypto and (2) tech crypto
If you can’t tell them apart, you’ll stay hopelessly confused
foufrix retweeted
🔥 INTRODUCING THE ODHACK WEEK! 🔥
Welcome to the very first edition of the ODHack where open issues are up for grabs and 100K grants ($, OP & STRK) will be distributed 🥳!
👀We've all been there—looking through countless repos, searching for open issues that match our skill sets...
✨The goal is therefore to connect you with AMAZING projects who need your contributions NOW and in return, reward you for a job-well done!
🫡 So don't miss out, GO, GO, GO! YOU HAVE TILL 25TH OF FEB.
🫶 We are so thankful to start this edition with @gizatechxyz @ohayo_dojo @nethermind @argentHQ BlobStream @Starknet @0xFableGame @AvailProject !!
🙏 Big thank you to the project leads who helped get the ball rolling @tarrence @DrSpacemn @tjelailah @raphael_dkhn @norswap @robin_rrtx @abdelstark & @asparenb!
All info here 👉blog.onlydust.com/blog/odhac…
And P.S. there will be merch ... 😉
Farcast.xyz is interesting.
User own their data but have to pay for it, have unlimited access, no ads (5-7$ annual, twitter pro 168$)
Dev side, anyone can run a node (called Hubs) and query as much as they want. Twitter API is 100$-10k GET - 5000$-1M GET per month
1
3
385



























