OpenAI introduced a testing tool called EVMbench in their post. This tool is specifically designed to test the ability of AI agents to find bugs, exploit bugs for destruction, and fix bugs on smart contracts. Smart contracts are automated programs that run on the chain, such as those on Ethereum, which lock billions of dollars in encrypted assets and can easily be stolen by hackers if there are vulnerabilities. Why is OpenAI doing this? The main purpose is to demonstrate the potential of AI in the field of blockchain security, helping developers to audit and strengthen contracts faster, and reducing economic losses in the real world. At the same time, they also showcased that their latest model GPT-5.3-Cdex performed well in this test (such as successfully exploiting 72% of vulnerabilities), showing improvement compared to the old version six months ago. So, what impact does EVMbench have on Ethereum? Overall, it is a positive signal that is conducive to promoting a safer and more stable Ethereum ecosystem, although there are also negative aspects. Let's break it down specifically: Positive aspects: Strengthening security audits, promoting defense upgrades, and promoting ecological benefits. In terms of security auditing, there are a large number of DeFi projects running on Ethereum, locking billions of dollars in encrypted assets, making it easy for hackers to steal money at the slightest vulnerability (more than 3.4 billion dollars in encrypted assets will be hacked in 2025 alone). EVMbench enables AI to automatically detect, exploit, and fix vulnerabilities, which can help developers audit contracts faster and cheaper. Previously, some manual audits took weeks and tens of thousands of dollars, but now AI can try a round in just a few minutes, making it more efficient. Of course, high-end complex contracts still rely on manual review by human experts and the imaginative "hacker thinking". In the long run, there is an opportunity to reduce real-world hacking losses, and with the widespread use of such tools, similar incidents may be reduced. • Promoting defense upgrades: OpenAI says their focus is on defense, and they also plan to open source tools and provide funding to encourage more people to use AI to strengthen contracts. This is a benefit for Ethereum developers, as it can make the entire network more resistant to attacks. From the results, it can be seen that their latest model GPT-5.3-COdex has a success rate of 72% in exploiting vulnerabilities. Ecological benefits: As the largest smart contract platform, Ethereum, if AI auditing becomes a standard feature, can attract more projects and funds to come in, enhancing overall trust. Some even say that this will accelerate the automation of Web3 security and make DeFi more reliable. Negative aspects: increased attack risk+double-edged sword effect. • Increased attack risk: AI is particularly strong in "exploiting vulnerabilities to steal money", and if someone uses such a tool, it can be easier to find and attack contracts. After each newly deployed contract goes online, it may face constant scanning and attack attempts from AI. The double-edged sword effect: Although OpenAI emphasizes defense, AI technology is neutral and anyone can use it. In the short term, the attacking side may have an advantage because the attack only requires "stealing once" and the defense is comprehensive.