More evidence your AI agents can be turned against you

Aikido researchers disclosed a prompt-injection vulnerability impacting major AI coding assistants integrated into development automation (e.g., GitHub Actions). By embedding commands in repository content (commits, issues, PRs), attackers can cause LLMs to treat that content as instructions, potentially executing shell commands, editing PRs/issues, and leaking privileged GitHub tokens; vendors have been notified and some fixes applied, but the root cause is architectural and may affect many projects.