AI auditing is here OpenAI has officially launched EVMbench! ************Bookmark this************ A benchmark test developed in collaboration with Paradigm, based on 120 high-severity vulnerabilities (many from Code4rena competitions), to evaluate the three major capabilities of AI agents on EVM smart contracts: - Detect: Identify vulnerabilities (recall rate is decent, but often misses multiple issues) - Exploit: Successfully exploit (GPT-5.3-Codex achieved 72.2%, a huge leap from GPT-5's 31.9% six months ago) - Patch: Fix and defend (low success rate, often breaks functionality after patching) Using a Rust-based harness to ensure safe and deterministic testing, simulating a sandbox environment to replay transactions. For DeFi security, this is a double-edged sword: AI accelerates vulnerability discovery (↑ attacks) but also enhances automated auditing (↑ defense), especially in protecting assets worth $100B+. Now, let’s talk about the auditing industry: Will this disrupt traditional human auditing? Human auditors have the risk of misconduct—think about how many vulnerabilities are discovered but kept private (the temptation of blackhat behavior is high), or missed due to negligence/bias. AI, on the other hand, is objective, motive-free, and purely data-driven. However, AI is still weak in detect/patch capabilities and cannot fully replace human domain knowledge and creative judgment yet. Prediction: In the short term, AI will become an auxiliary tool (e.g., speeding up preliminary audits). In the long term, if it evolves to be production-ready, it could reshape the industry, making auditing more efficient and transparent. Those vulnerabilities that were "kept hidden"? At least AI won’t keep secrets The on-chain world is entering the AI × EVM era. Do you think the auditing industry will be disrupted? Bullish or bearish? https://(openai.com)/index/introducing-evmbench/ EVM SmartContract AIAgent DeFiSecurity Web3 BlockchainAudit