This is a rare case where a major company's security research was turned into something the broader community can actually run and inspect. Both the Cloudflare post and the audit repo are worth…

Mythos could stitch findings into full exploit chains and generate working PoCs across 50+ repositories. Other models found individual bugs but often failed at the "putting it all together" step. The…

𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗪𝗼𝗿𝘁𝗵 𝗥𝗲𝗮𝗱𝗶𝗻𝗴 - 𝗪𝗲𝗲𝗸 𝟮𝟭, 𝟮𝟬𝟮𝟲 A great week to look into AI-assisted tooling ⚒️ 𝗲𝘃𝗶𝗹𝘀𝗼𝗰𝗸𝗲𝘁 / 𝗮𝘂𝗱𝗶𝘁 An 8-stage vulnerability-discovery agent based…

Agent infra is moving from “chat with code” to: - endpoint security - local model workflows - narrow agents with real pipelines Trending tells you what people have already noticed new + fast tells…

通过 8 阶段多 Agent 协作流水线，利用 Claude 订阅自动发现代码库中的安全漏洞 项目复刻了 Cloudflare Project Glasswing 的漏洞发现架构，用 8 个不同阶段的 Agent 分工合作：先扫描代码结构，再按攻击类别逐个探测，然后用不同模型反驳验证，证明攻击路径可达，最后生成结构化报告。走 Claude 订阅计费而不是 API Key，也支持…

Most "AI security" tools are one prompt: find bugs. evilsocket's audit runs 8 narrow stages. Second agent disproves every finding. Reachability trace checks if attacker input reaches the sink. → 308…

Earlier today Cloudflare's CSO shared how they tested Anthropic Mythos using an unreleased 8-stage vulnerability-discovery agent. So I asked Opus to implement the agent for me, it works via Claude…