AGENTS.md is the new executable: sites can drop agent instruction files into your Downloads folder
alvelda · x · 2026-11-19
Forever Security documented a new attack against AI coding agents, caught live on a real customer's machine. Claude Code and Codex automatically read and follow instruction files like AGENTS.md in whatever folder they open — so a website can silently drop such a file into Downloads just by being visited, tricking the agent into handing over credentials or running attacker code. No click or user mistake is needed, and it can't simply be patched since it exploits intended behavior. Roughly half of measured companies had employees running these agents, and every browser user among them is exposed. Advice: instruction files are executable context — never start an agent in a folder you didn't curate.
More from coding & agent
- Switch Buddy builds a full dev team from one prompt: coder, reviewer, debugger in shared rooms — cneuralnetwork · 2026-11-19
- Unverified: Google's rumored Gemini Neon wows with one-shot working Olivetti typewriter — dreamwieber · 2026-11-19
- Zed dev: Gemini 4 Neon solves in one round what 3 frontier models couldn't crack in 9 — vivnat · 2026-11-19
- AUSM: open-source local-first context layer for context-aware macOS apps — HaijunXia · 2026-11-19
- Early tester: Neon 'abnormally good' even before release, watching the model build is stunning — AI_Andrew · 2026-11-19
- System 1 models as harness tools with codemode are underrated, dev says — GabGarrett · 2026-11-19