According to WuShuo, a new research report released by a16z crypto reveals that their research team conducted systematic tests on the ability of AI agents to identify and autonomously exploit vulnerabilities in the DeFi space. In baseline tests without prior knowledge, the agents achieved an initial success rate of 50% by 'cheating'—using the Etherscan API to retrieve post-attack transaction records. When structured domain knowledge (such as vault donations, AMM balance manipulation, etc.) was injected into the agents, the success rate increased to 70%, but still fell short of 100%. a16z concluded that while AI agents can currently assist in identifying vulnerabilities and generating simple PoCs, they are still unable to replace professional security experts when dealing with complex, multi-step economic exploits. https://www.(wublock123.com)/news/news-60281