Ethereum Foundation says AI bots find bugs but struggle with accuracy

The Ethereum Foundation is letting the machines take a crack at network security. The Protocol Security team recently deployed a swarm of coordinated AI agents to hunt for vulnerabilities in core infrastructure, but the experiment is yielding mixed results. While the bots successfully flagged legitimate flaws, they’re still drowning human researchers in a sea of false positives. It's a classic case of quantity over quality. The team found that while LLMs can scan code at a scale humans can't touch, they lack the nuance to distinguish between a catastrophic exploit and a harmless quirk in the logic. One specific test run showed that AI excels at identifying patterns but fails when faced with the idiosyncratic architecture of the Ethereum Virtual Machine. These digital auditors won't be replacing human bug hunters anytime soon. They're currently more like over-eager interns than seasoned security pros. Can these models learn to filter their own noise before the next major hard fork?