evmbench
171 skatījumi
4 piedalās diskusijā
Populārie
Jaunumi
Skatīt tulkojumu
AI vs. Hackers: Measuring Smart Contract Security with EVMbench
OpenAI and a crypto company called Paradigm just released something called EVMbench. It is basically a testing tool like a benchmark to see how good AI models are at dealing with security problems in smart contracts the code that runs on blockchains like Ethereum and holds over 100 billion in crypto.
Smart contracts can have bugs that let hackers steal funds so this benchmark checks if AI can help fix that risk. They built it using 120 real vulnerabilities pulled from 40 actual professional security audits including some from a blockchain project called Tempo.
The test has three main parts:
• Detect: Can the AI spot the bugs when it looks at the code
• Patch: Can it fix the bugs without breaking how the contract is supposed to work
• Exploit: Can it carry out an attack to drain the funds in a safe test environment not real money
They ran some top AI models on it. The newest one from OpenAI called GPT5.3Codex did really well at the exploit part it succeeded 72.2 percent of the time. That is a big jump from the earlier GPT5 which only managed about 31.9 percent six months ago.
The idea is to keep an eye on how fast AI is getting better at both finding and fixing these issues which is good for security and also at exploiting them which shows new risks we need to watch out for. They say it is important to use AI more for defense like having it audit and strengthen contracts to stay ahead of potential problems. Overall it is a way to measure progress in this area and make blockchain stuff safer as AI gets smarter.
$ETH
#OpenAI
#AI
#EVMBench
#Web3
$AI
Smart contracts can have bugs that let hackers steal funds so this benchmark checks if AI can help fix that risk. They built it using 120 real vulnerabilities pulled from 40 actual professional security audits including some from a blockchain project called Tempo.
The test has three main parts:
• Detect: Can the AI spot the bugs when it looks at the code
• Patch: Can it fix the bugs without breaking how the contract is supposed to work
• Exploit: Can it carry out an attack to drain the funds in a safe test environment not real money
They ran some top AI models on it. The newest one from OpenAI called GPT5.3Codex did really well at the exploit part it succeeded 72.2 percent of the time. That is a big jump from the earlier GPT5 which only managed about 31.9 percent six months ago.
The idea is to keep an eye on how fast AI is getting better at both finding and fixing these issues which is good for security and also at exploiting them which shows new risks we need to watch out for. They say it is important to use AI more for defense like having it audit and strengthen contracts to stay ahead of potential problems. Overall it is a way to measure progress in this area and make blockchain stuff safer as AI gets smarter.
$ETH
#OpenAI
#AI
#EVMBench
#Web3
$AI
OpenAI un Paradigm uzsāks AI aģenta rīku viedo līgumu drošībai
#OpenAI ir sadarbojies ar #Paradigm , lai uzsāktu #EVMbench , atsauces punktu, kas mēra, cik labi AI aģenti var identificēt, labot un izmantot kritiskas viedo līgumu ievainojamības. EVMbench balstās uz 120 reālām ievainojamībām, kas pārbaudītas 40 auditos, tostarp scenārijos no #Tempo blokķēdes audita, kas koncentrējās uz augstas apjoma stabilo monētu maksājumiem.
EVMbench mēra trīs galvenās spējas. Detektēšanas rādītājs nosaka, cik labi aģenti identificē zināmās viedo līgumu kļūdas. Labojums novērtē, vai aģenti var novērst problēmas, nepārkāpjot paredzēto uzvedību. Izmantošana pārbauda aģentu spēju veikt pilnīgas līdzekļu izsūknēšanas uzbrukumus izolētā blokķēdes vidē.
👉 openai.com/index/introducing-evmbench/
#OpenAI ir sadarbojies ar #Paradigm , lai uzsāktu #EVMbench , atsauces punktu, kas mēra, cik labi AI aģenti var identificēt, labot un izmantot kritiskas viedo līgumu ievainojamības. EVMbench balstās uz 120 reālām ievainojamībām, kas pārbaudītas 40 auditos, tostarp scenārijos no #Tempo blokķēdes audita, kas koncentrējās uz augstas apjoma stabilo monētu maksājumiem.
EVMbench mēra trīs galvenās spējas. Detektēšanas rādītājs nosaka, cik labi aģenti identificē zināmās viedo līgumu kļūdas. Labojums novērtē, vai aģenti var novērst problēmas, nepārkāpjot paredzēto uzvedību. Izmantošana pārbauda aģentu spēju veikt pilnīgas līdzekļu izsūknēšanas uzbrukumus izolētā blokķēdes vidē.
👉 openai.com/index/introducing-evmbench/
Skatīt tulkojumu
OpenAI + Paradigm Drop EVMBench – AI Shield for Crypto Tokens & Smart Contracts!
Game-changer for Web3 security! OpenAI just launched EVMBench, a cutting-edge benchmarking system with Paradigm to test AI agents on detecting, exploiting, and patching vulnerabilities in Ethereum smart contracts & tokens.
Core Features
EVMBench evaluates AI across three modes in a sandboxed EVM:
• Detection: Spot flaws in Solidity code (tokens, DeFi).
• Exploitation: Simulate real attacks.
• Patching: Auto-fix issues with explanations.
Open dataset of real/vulnerable contracts + 10M$ OpenAI cybersecurity fund. Aims to slash billions in hacks via standardized AI audits.
Impact
Paradigm calls it "open benchmark revolution" no more manual audits scaling issues. OpenAI's first crypto-specific tool post-EO clarity. DeFi safer, faster launches ahead!
Bullish for AI-blockchain fusion? Thoughts?
#OpenAI #EVMBench #SmartContractSecurity #CryptoAi #paradigm
$ETH
Core Features
EVMBench evaluates AI across three modes in a sandboxed EVM:
• Detection: Spot flaws in Solidity code (tokens, DeFi).
• Exploitation: Simulate real attacks.
• Patching: Auto-fix issues with explanations.
Open dataset of real/vulnerable contracts + 10M$ OpenAI cybersecurity fund. Aims to slash billions in hacks via standardized AI audits.
Impact
Paradigm calls it "open benchmark revolution" no more manual audits scaling issues. OpenAI's first crypto-specific tool post-EO clarity. DeFi safer, faster launches ahead!
Bullish for AI-blockchain fusion? Thoughts?
#OpenAI #EVMBench #SmartContractSecurity #CryptoAi #paradigm
$ETH
Pieraksties, lai skatītu citu saturu