Newest papers

Honeypots for AI Agent Insider Threats

Jan Wehner, Jam Kraprayoon · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.26664 · cited by 0 works (OpenAlex) · open-access PDF

Hindi

संगठनों के भीतर तैनात एआई एजेंट एक आंतरिक खतरा पैदा कर सकते हैं यदि वे गलतियां करते हैं, एक हमलावर द्वारा संकोच किए जाते हैं, या असंगठित लक्ष्यों का पीछा करते हैं। हम एआई एजेंट तैनाती की सुरक्षा में सुधार करने के लिए एक एआई नियंत्रण विधि के रूप में हनीपॉट्स का प्रस्ताव करते हैं। केवल तैनाती से पहले मूल्यांकन के लिए हनीपॉट्स का उपयोग करने के बजाय, वकीलों को सुरक्षा टीमों को हानिकारक कार्रवाई के बारे में विश्वसनीय चेतावनी प्रदान करनी चाहिए, खराब व्यवहार का पता लगाने के लिए, हानिकारक एजेंटों को दोषी ठहराने के लिए, और हमलों को रोकने के लिए। ऐसी हनीपॉट्स वास्तविक तैनाती

Machine translation (Meta M2M100); the English original follows.

Abstract

AI agents deployed inside organizations can pose an insider threat if they make mistakes, are compromised by an attacker, or pursue misaligned objectives. We propose honeypots as an AI control method to improve the security of internal agent deployments. Rather than only using honeypots for pre-deployment evaluations, defenders should plant decoy opportunities for harmful actions throughout the production environment to detect misbehavior, incriminate malicious agents, and deter attacks. Such honeypots can elicit misbehavior under real deployment conditions, identify malicious agents even if chain-of-thought monitorability degrades, provide reliable warnings about agent misbehavior to security teams, and scale cheaply to many agent instances. The presence of honeypots also introduces uncertainty about which opportunities are real, thus deterring malicious agents from attacking and reducing the rate of misbehavior. We discuss the limitations of honeypots and give reasons to expect them to remain effective as agent capabilities increase. Finally, we outline key research questions to measure and improve the effectiveness of honeypots: testing the robustness of honeypots, comparing their detection effectiveness with other monitoring methods, understanding their incrimination and deterrence effects, and building supporting infrastructure.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Honeypot, Computer security, Insider, Computer science, Software deployment, Deterrence theory

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange