Threat
In-the-Wild Attack Abuses AI Agents' Trust in Markdown
First reported darkreading.com
Page published
Earliest dated coverage: 26 Oct 2026 · First observed: 26 Oct 2026 · Latest dated coverage: 26 Oct 2026
Coverage timeline
Single-source incident — one report is available.
Why it matters
The AGENTS.md attack demonstrates how downloaded Markdown files can cross an AI coding assistant's trust boundary, turning untrusted local content into instructions for potentially harmful actions.
Forever Security reported an in-the-wild attack attempt involving a malicious AGENTS.md file in a customer's Downloads folder, intended to make an AI coding assistant follow attacker-controlled instructions to install malware and steal credentials. The company said it blocked the attempt before any code executed.