yudan.eth retweeted
Use our tx debugger agent skill to understand the chain of events in this exploit ⤵️
➡️ Throwaway contract freed ETH-A collateral from 4 GEB SAFEs via GlobalSettlement
➡️ exited it as WETH
➡️ unwrapped it
➡️ forwarded ~5.9436 ETH back to the attacker
nitter.cf/exvulsec/status/209481…
🚨 ALERT — Exploit on Ethereum
A GEB/RAI-style CDP deployment (Reflexer's GEB framework, in Global Settlement since Jan 2021) was just drained of its leftover ETH-A collateral. ~5.94($14k) ETH stolen.
Root cause:
several SAFEs were owned by the shared GebProxyActions library itself (0x84fe452d9fb495a335c74a225e6ad52c35eb8616), not by user proxies. Its quitSystem() is public and unauthenticated, so the attacker called it directly — GebSafeManager's msg.sender == ownsSAFE[safe] check passed (the library calling as itself) — and migrated other users' collateral to himself, then freeCollateral + exit.
Flow (per SAFE): processSAFE → quitSystem → freeCollateral → CollateralJoin.exit → WETH.withdraw
Attack tx: etherscan.io/tx/0xfbce28e35c…
Attacker: 0xb929c7215c0ec8ebad5fbf73b1da63bccfff1896
Exploit contract: 0x6a213f0b5bd9eed865d3e2efc867b73dfe9039e7
Lesson: never let a stateless, shared "proxy-actions" contract become the registered owner of a position — its public helpers turn into anyone's withdraw button.
Subscribe to our TG bot for real-time attack alerts 👉 exvul.com/skyeye-monitor
今晚别刷Netflix了。
去看这个斯坦福2小时34分钟的讲座。
这是我见过最清晰、端到端讲透ChatGPT和Claude到底怎么建出来的课。
从Tokenization和BPE开始,一路到Transformer架构、训练管线、Next-token解码。
不管你是从没写过一行AI代码,还是天天在ship Agent,看完都会把很多“原来如此”的点一次打通。
真正核心就这几层:
1文本怎么变成模型能吃的数字(BPE tokenization)
2语言模型本质就是在预测下一个token
3Transformer怎么用Attention做消息传递
4训练时用NLL loss把整条序列概率拉高
5生成时怎么一步步解码出来
大多数人花好几年才拼出的全景图,这节课直接给你画完。
把时间空出来,这可能是你这个月最值的一堂课。
yudan.eth retweeted
zk-skills by @zksecurityXYZ was just added to pashov/ai-web3-security🤝
Bravo to the zkSecurity team for shipping this👏
github.com/pashov/ai-web3-se…
yudan.eth retweeted
If Vitalik could accept a staking-without-reward solution, i would agree.
But if Vitalik is so divinely enlightened,
why doesn't he just donate all his ETH assets to the community?
This is a paradox of power and responsibility.
You can't expect anyone to work for you without receiving any salary, and then be lectured "for the future of Ethereum."
Please, this kind of utopian communist rhetoric is a trick to fool children.
Ethereum is a system of fundamental, real technological innovation and asset models, not a Paris Commune game.
The author who proposed this EIP is either an idiot or a complete idiot.
🚨 New EIP: Tapered Issuance Burn
We just submitted an EIP to ethereum/EIPs: a minimal, market-driven fix to Ethereum's issuance policy removing the incentive for stake growth beyond 50% of ETH supply.
EIP-8361 by @pintail_xyz, @jdetychey, @dapplion, @pa7x1, @ladislaus0x & @drakefjustin 🧵
完全不切实际的提案,和一开始EIP1559的初衷背道而驰。ETH本身不是BTC,这样做只会削减个人验证者的积极性
以太坊新提案EIP-8361:质押率超50%后收益直接归零
以太坊研究员Jerome de Tychey等六人联合提交新提案EIP-8361,设计一种"渐进式发行销毁"机制,让质押率超过50%后共识层收益归零。
目的是消除无限扩张质押的动力(现有机制下,就算全部ETH质押,年化收益也不会低于约1.5%)。收益下调会分18个月逐步过渡,再加约6个月分叉准备期。
不过Aave创始人Stani Kulechov公开质疑,称这一方案会让质押收益变得不可预测甚至不划算,可能反而削弱以太坊、把散户验证者挤出局。
1
2
232
yudan.eth retweeted
Our EVM bytecode decompiler is live.
Address in, readable Solidity out. Unverified contracts too.
Ethereum + BNB Chain. No login — just open it.
Enjoy 🫡 decompiler.exvul.com
🚨Solidity Devs - this awesome repository has 50 agents, 91 skills and 19 hooks for Solidity development, deployment, testing and security
Built & compiled by a Uniswap Senior Engineer. Use this for higher quality development, MIT license🫡
github.com/ccashwell/evm-cor…
yudan.eth retweeted
Kimi K3 exploited the latest Redis server with a 0day it discovered.
All it took was 27min with 32 agents.
github.com/berabuddies/redis…
当年也是有个链游,快倒闭了 7/13 多签被黑,现在想来或许是为了光明正大分钱而已
Blockaid detected an exploit at 2026-07-22 21:30 UTC targeting @AFX_XYZ, a protocol on @arbitrum. The exploit was specific to a bridge that AFX operates. Approximately 24.15M USDC has been drained thus far from the protocol.
Our team has been working with the incredible folks on the Arbitrum team to respond to the incident, to engage with the affected protocol, and to help them contain the stolen funds.
Exploit tx: arbiscan.io/tx/0x50d0b3ec6c3…
164
如果是 benchmark验证,为什么要留下一个特殊的网络接口在外面?
We're partnering with @huggingface to investigate an unprecedented security incident.
Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation.
Sharing preliminary findings to help defenders understand emerging risks:
openai.com/index/hugging-fac…
189
yudan.eth retweeted
A massive, well-organized database of real on-chain exploits with full technical breakdowns.
- 845+ reproduced incidents
- detailed analysis for every hack
- Foundry PoCs you can run immediately
- easy navigation by chain and vulnerability type
crypto.training/hacks/
We just spent $644 to benchmark @pashov with Opus 4.7 MAX:
Opus 4.7 — 48/94 found · $2.18/finding
Pashov skill — 49/94 found · $11.01/finding
Same model underneath, 5× the spend for one extra finding.
Anyone having similar results? Dataset in comment.
This is a very typical case of evolutionary stagnation, or even reverse evolution.
For this problem, I have already found a solution and will publish an article about it soon. Welcome to read it
We just spent $644 to benchmark @pashov with Opus 4.7 MAX:
Opus 4.7 — 48/94 found · $2.18/finding
Pashov skill — 49/94 found · $11.01/finding
Same model underneath, 5× the spend for one extra finding.
Anyone having similar results? Dataset in comment.
yudan.eth retweeted
🚨Community Alert:
Blockaid Exploit Detection system detected an exploit involving the @jaredsmev MEV bot on Ethereum.
The incident resulted from attacker-controlled contracts tricking an automated MEV execution system into granting token approvals, later used to drain funds.
7.5M$ drained.
Exploit tx:
etherscan.io/tx/0x2be8704f5a…
More details in🧵
yudan.eth retweeted
先不管说Drift说的朝鲜黑客是不是事实
在现今这个周期,DeFi项目方监守自盗基本上早就是过去时了。你能看到的超过2000万美金以上的被盗事件,基本上都是“真的被盗”
我一线参加过多次重大DeFi安全事故善后,整个善后过程其实是非常繁琐的:
- 事故发生当下,需要你在极为有限的时间内(通常只有几十分钟),定位事故原因,采取紧急措施尝试终止资金失窃
- 同一时间,你所在的生态、CEX、上下游合作方、用户、公域社媒都会轰炸你的各种联系频道,你还需要第一时间去通气和安抚
- 然后这时候需要联系相当多的执法部门:包括但不限于Homeland security,IRS,各国警务,和专门处置这种安全事故的安全公司、跨链桥、稳定币发行方等等
- 在这个时间段里,你基本上是无法对外、尤其是对公域保持信息同步的。一是法务上根本不允许,二是你尚处在攻防战博弈(比如试图看看能不能联系攻击者接受白帽)阶段,根本不可能对外透明
这也就是为什么这个阶段很多项目方挨骂,被人说监守自盗是没办法自证回应的原因
等到所有的这些全部过去,才能开始给出post motem,对社区和生态开展通气会、解释事故来龙去脉,并且制定赔偿善后方案
注意:以上说的这些,是在非常短的时间内发生的。如果团队之前没有经验,或者没有处置预案,是根本应付不来的
指望除了CEX之外的三方基本上是指望不上的,比如说 @zachxbt 说的USDC的问题——所有人都是众人自扫门前雪,尤其是这些北美“合规”基建,光法务流程就要走好几个月。要是等他们解决问题,黄花菜都凉透了
至于监守自盗,这么多双眼睛盯着,而且执法部门第一时间就一定会先assume监守自盗,所以介入第一时间就需要把所有的材料进度共享,根本没有余地(当然离职后盗窃这种是另外一回事)自己rug。
如果要rug,有更多更聪明的方式去做,不可能搞得这样全网关注
对于安全,其实还有几个实操后非常反直觉的认知:
1. 越是出这样的事故,就越会催动像私有链perp "DEX“ 的中心化方案出现,比如像Hyperliquid或者一些其他的私有链友商,本质其实是分布式服务器结构、不KYC和审核提币的CEX
链上DAPP级别的交易所被盗,是无法卡住资金流出的。攻击者可以通过跨链桥等快速的把资金洗出去
但是私有链本质上就是中心化服务器,节点卡住=拔网线,资金完全出不去,属于终极解决方案
很多这类项目方包括中心化交易所,当年都是向自己的大LP和MM展现了敢于果断拔网线承担社区骂名的风险,才获得信任和大规模增长的
当然这里的取舍就是:你觉得项目方作恶几率大,还是朝鲜黑客攻击几率大
2. 全开源项目安全压力更大:
如果这个协议是全开源的,那么相当于你在明,攻击者在暗。人家有的是时间和信息来研究如何攻击,而绝大多数的开源DeFi都不可能像Uniswap和AAVE一样由全网fork,大家共同维护;而是依靠一个团队一个公司维护,开源的好处没得到,坏处一样不少
攻击事件发生后,很多协议都会被指责不开源才导致攻击,但事实上如果不开源都被攻击成功,那么开源的只会出事的更快
3. 英语系项目方危险程度更高
别误会,这里不是给自己作为华人项目方脸上贴金 —— 谁有技术漏洞都是死。而是说英语系项目方的那种招聘和开源协作的形式更容易遇到朝鲜黑客这样的供应链攻击。
华人项目一般都是封闭开发,技术岗不是信得过的人不招。但是英语系国家不是这样的,有的时候说开源共建说的自己都信了,忽略了黑暗丛林法则的真实存在
最后,希望Drift团队能想办法挽回损失吧
yudan.eth retweeted
Claude Code 源码泄露,不到 24 个小时。
claw-code 成为 GitHub 历史上最快突破 3 万 Star 的开源项目,还在持续暴涨中。
项目作者在发现源码泄露后,选择从零开始用 Python 重写,实现了 Claude Code 的核心功能。
目前作者已启动 Rust 移植工作,运行速度更快、内存更安全,计划作为该项目的最终版本。
GitHub:github.com/instructkr/claw-c…
感兴趣的朋友可以持续关注一下。
Trail of Bits, the OG Tier 1 web3 security company just crushed it with a NEW SKILL on Logic Issue Detection
Beautiful contributions, this is how you do it in web3 security guys🫡Thank you @thebensams & @dguido👏
Replying to @trailofbits
@trailofbits 's dimensional-analysis is *provenly the best auditing skill to detect logic vulnerabilities in smart contracts
you may disagree - as I only compared it to 2 other logic-focused bug finding skills, and only ran a couple of iterations against each - but that's the best benchmark out there for this so far
best part - you just need 1 click and an AI agent to contribute to the benchmark and try it yourself
forefy.com/benchmarks/058b23…































