Angehefteter Tweet
Dan Tehrani
942 posts

Dan Tehrani
@dan_tehrani
security researcher prev co-founder @personae_labs, @Mikawaya9
SF/Tokyo Beigetreten Mart 2020
502 Folgt797 Follower
Dan Tehrani retweetet
Dan Tehrani retweetet

1/ “Who is the top JIT liquidity searcher in Ethereum?” → a question answer by our paper w/ @0xAlphaist (accepted to a FC workshop: fc26.ifca.ai/defi/)
We mapped JIT activity in Uniswap v3 and linked searchers <-> builders from Jan 2024 to Sept 2025.
One name dominates: jared accounts for >50% of all observed public JIT execution.

English

Latest version of Prettier is broken on Cursor. This GitHub issue has the fix (just force downgrade to v11.0.2): #issuecomment-3761391774" target="_blank" rel="nofollow noopener">github.com/prettier/prett…

English

Our LLM auditor has identified a Medium severity vulnerability in a bug bounty as well!!
More updates soon
x.com/dan_tehrani/st…

Dan Tehrani@dan_tehrani
I’ve been building an AI agent for finding vulnerabilities in smart contracts. It has found several confirmed bugs in audit contests, including code4rena contests which codebases are pre-scanned with @zellic_io's v12 LLM auditor. Tuning the prompts of an AI agent to find vulnerabilities requires thousands of dollars in inference cost for the best reasoning models, and careful try-and-error prompt tuning across many real-world vulnerabilities. It still falls short compared to the best human auditors, but with the prompts and the base models improving, the gap will only get smaller.
English

We're offering *free* audits with our LLM auditor.
Please DM if you want to get an audit report now.
Our system usually costs hundreds of dollars per audit for LLM inference. It has been competing in audit contests and has identified variaous high/medium severity vulnerabilities.
We currently provide *free* audits for projects in exchange for permission to write a case study of auditing your project.
Please DM if you are interested:)
x.com/dan_tehrani/st…


English

I’ve been building an AI agent for finding vulnerabilities in smart contracts.
It has found several confirmed bugs in audit contests, including code4rena contests which codebases are pre-scanned with @zellic_io's v12 LLM auditor.
Tuning the prompts of an AI agent to find vulnerabilities requires thousands of dollars in inference cost for the best reasoning models, and careful try-and-error prompt tuning across many real-world vulnerabilities.
It still falls short compared to the best human auditors, but with the prompts and the base models improving, the gap will only get smaller.



English

@HBBHZjW4TU-_rDF7GQh3lg/H1CR1mbZZe" target="_blank" rel="nofollow noopener">hackmd.io/@HBBHZjW4TU-_r…
ZXX

In the last few months, I’ve been focusing on building an LLM-based smart contract auditing system with the question: Can an AI auditing system become more reliable than human auditors?
I see a lot of sentiment that discards AI auditing.
So I wanted to assess what AI actually can and cannot do in finding vulnerabilities.
Here are all my learnings (post link in thread)
With today’s models, LLMs aren’t capable of finding all vulnerabilities expert humans can find with just a few prompts. But with the base model improving and further research in promoting, it’s plausible that LLMs will become more reliable than humans.
Further research is required, but as millions of dollars are at stake, we need to keep pushing the frontier of LLM auditing faster than the blackhats do.
English

@PraveenInPublic this is how I approach learning something complex
English

@robert_shaw don't know why but I have an urge not to click it whenever I see ✨
English

@Harish_52 A tool to engage more on X without losing your mind
x.com/dan_tehrani/st…
English

@devfaizanali A tool to reply more on X, but without the chaos of X
x.com/dan_tehrani/st…
English









