Researcher builds hardest-ever RCE chain largely AI-driven with GLM
banteg · x · 2026-11-09
Security researcher makemakekbo says his long-teased RCE exploit chain is nearing completion — the most technically challenging thing he has ever built, and much of it was AI-driven using GLM. He could fan out and test many hypotheses in parallel but still had to drive the process, as models can't yet separate good lanes from bad. His takeaway: a frontier LLM a year from now could likely drive the whole chain end-to-end, likening today's models to 80s-90s chess engines — deep search but 'stupid' — until they beat every human. Screenshot shows vtable/PLT-stub exploitation details.
More from coding & agent
- Running 35 Local Agent Repos: launchd Calls Stale Copies Until You Pin Absolute Paths — petrusenko_max · 2026-11-09
- Dev vibe-codes a drag-to-inspect UI element tool with Claude, open-sources it under MIT — jasonkneen · 2026-11-09
- Facet launches a touch-first Mac browser with Claude Code and Codex agents in the sidebar — dean_rie · 2026-11-09
- Critical unauthenticated RCE in React Server Components, patched in 19.x point releases — DanielLockyer · 2026-11-09
- David Sacks: Salesforce made Claude its front end — SaaS is turning into an agent database — 2C_ornot2C · 2026-11-09
- Building agents is hill climbing: early results are noisy and gains may not generalize — andreisavu · 2026-11-09