CoinWorld News: The Ethereum Foundation has recently published a technical article noting that the biggest challenge for AI-assisted security research is proving whether reported vulnerabilities are real, rather than looking for potential errors. The Ethereum protocol security team said that recent experiments show coordinated AI agents discovered genuine software flaws in systems that Ethereum relies on, but most of the effort goes into distinguishing valid findings from false positives. The team emphasized that AI agents should be viewed as tools for generating hypotheses, not decision-makers. To reduce unreliable findings, the Foundation deploys multiple AI agents to review the same software library, with each agent handling a different stage of the review process. The process starts with reconnaissance, gradually narrows the attack surface, and ultimately ensures that each report can be verified in production code.