Security researcher: patching vulnerabilities one by one is a losing game in the AI coding era
agstrait · x · 2026-11-18
Security researcher dadrian argues in a long essay that defenders can't keep playing whack-a-mole with vulnerabilities now that AI coding tools write code at massive scale and speed. Even if each bug looks microscopically familiar, the macro-level changes in scale, speed, and friction have transformed the game.
The essay criticizes security industry identity culture—careers built around the hacker ethos and surprising vulnerability discoveries (citing DEF CON's 25,000 attendees and Freakonomics-style surprise worship)—arguing it rewards cleverness in finding the unexpected rather than systematically reducing risk.
A supporting anecdote: PHP developers recently found a memory-corruption bug in Java stemming from hand-rolled assembly code—evidence that all code is legacy code, and layers of technical debt are now being detonated by scaled-up, sped-up tools. Zeynep shared the piece as a solid explanation of why we can't patch our way out of insecure software in the age of AI coding.
More from coding & agent
- Opus 5.5 ran an AI company for 7 weeks with just an HF account and 600 GPU hours — 50k downloads, 100+ projects — maksym_andr · 2026-11-18
- Mistral details Vibe as a coding agent for large codebases, with long-horizon security analysis demo — MistralAI · 2026-11-18
- Mistral Vibe desktop app launches for macOS and Windows, free Pro trial for EU builders — MistralAI · 2026-11-18
- AI models a scene from scratch in Blender with no assets, author says barely intervened — op7418 · 2026-11-18
- Kilo Desktop Goes GA: Run Many Coding Agents and Review Diffs in One Window — anacondainc · 2026-11-18
- Researcher dissects artex: an MCP-powered pentest toolkit, not an AI, despite FBI claims — weskambale · 2026-11-18