FlagThis - Latest Cybersecurity News

Agentjacking is a novel attack vector targeting AI coding agents, specifically Claude Code, via indirect prompt injection. Attackers embed hidden instructions within ostensibly "clean" GitHub repositories, typically within documentation or setup files. When the agent analyzes the project, it is manipulated into executing unauthorized commands, often fetching malicious payloads via DNS TXT records to bypass static code analysis. By leveraging the agent's autonomous CLI capabilities, attackers can achieve remote code execution (RCE) and establish reverse shells on a developer's local machine, bypassing traditional security boundaries through the agent's trusted execution context.