Eliminating DoS bugs with ChatGPT: profiling 3,000 code sites and building a resource ledger
jd_pressman · x · 2026-11-15
jdpressman details a workflow to empirically eliminate DoS bugs from a web server using ChatGPT: have it write a profiler to find every request-path location where an attacker-controlled resource could cause a DoS, test under load with deployment benchmarks, manually categorize 3,000 locations, then build a formal resource ledger proving no data structure can be pushed into a DoS state—letting the model find and patch remaining flaws. The attached screenshot shows ChatGPT triaging a low-severity SSHA salt issue with a concrete fix. The author argues formal verification's real value is shrinking the search space so red teams attack the most vulnerable spots first.
Related event: Using ChatGPT to profile and eliminate DoS vulnerabilities(2 posts)→
More from coding & agent
- Multiplayer human-agent workflows are the underexplored future of coding — diegobuildsai · 2026-11-15
- One Kyoto photo becomes a walkable 3D world via Atlas, Three.js and Claude — Scobleizer · 2026-11-15
- 21st.dev offers 12,000+ crafted React UI components you can feed straight into AI coding agents — ericspo29 · 2026-11-15
- Sandbox resource caps risk hobbling production, but FOSS benchmarks ease the worry — jd_pressman · 2026-11-15
- Two prompts, pure Web Components: agent reuses 14 UI SDK behaviors out of the box — evilrabbit_ · 2026-11-15
- Open-source 22-chapter course teaches you to build production AI agents by using an agent — adnan_hashmi · 2026-11-15