arXiv 2609.21081

Security researchers published a paper on September 17 (arXiv 2609.21081) detailing a new attack class called Loopjacking that bypasses human-in-the-loop approval in AI agent frameworks by showing a h…