Sep 2026· International Conference on Public Key Infrastructure and its Applications· pp. 1-8· 0 citations· 22 references
TL;DR
Verifier-enforced inline revocation folds semantic content analysis and behavioral state tracking into the certificate verification path itself, positioning inline revocation as a practical containment primitive for autonomous agents.
Abstract
Autonomous large language model (LLM) agents can now make decisions and act at high speed. When compromised, they can themselves act as adversaries; therefore, monitoring their actions and controlling their identities are essential. The Public Key Infrastructure (PKI) that governs these identities was never designed for adversaries that act in seconds. Conventional revocation through Certificate Revocation Lists (CRLs) and the Online Certificate Status Protocol (OCSP) is reactive: a compromised certificate remains valid from the time of compromise until detection and status propagation, a gap of minutes to days. We close this detect-to-enforce gap with verifier-enforced inline revocation, which folds semantic content analysis and behavioral state tracking into the certificate verification path itself. When an agent's request is detected as malicious, the verifier revokes that agent's certificate. At the same time, the authoritative Certificate Authority (CA) reviews the case asynchronously. As a softer intermediate response, cryptoagility recredentials suspicious but unconfirmed agents, forcing re-authentication. Detection combines three components: a fine-tuned Modern-BERT classifier, a stateful behavioral predictor that tracks intent escalation across requests, and a five-score anomaly fusion engine that drives a multipathway revocation decision. Across 1,000 simulated agents (500 benign and 500 malicious) under four attacker policies, the fused pipeline achieves 0.942 F1 and 98.6% agent-level recall. Notably, 27.4% of malicious agents are revoked on their first request at 97.2% precision, and 86.4% within five requests. These results position inline revocation as a practical containment primitive for autonomous agents.
Mobile agents built with large language models (LLMs) act with user-granted authority while processing content from advertisements, messages, and third-party applications. Studies show that such content can redirect agents into attacker-chosen workflows, but do not isolate how earlier untrusted content can influence an...
Mahmoud Nazzal, M. Rabiu, Yue Xiao· Proceedings of the AAAI Symp...· 0 citations
This study runs a deterministic, checklist-based semantic validation pipeline that emits signed binary verdicts and commits rationale hashes on-chain, so any verdict can be replayed off-chain while on-chain settlement stays minimal.
Ayman Nait Cherif, M. Youssfi, O. Bouattane· IAES International Journal o...· 0 citations
This work proposes a framework based on the principles of capability based security with decentralized agent identity whereby agents access services based on tokens that are cryptographically bound to the agent's and issuer's identities and specify their scope.
Srikumar Subramanian, S. Sengupta· arXiv.org· 0 citations
Autonomous AI agents are increasingly used as enterprise identities with access to OAuth tokens, API keys, and delegated permissions across cloud and SaaS environments. However, existing Identity and Access Management (IAM), Privileged Access Management (PAM) frameworks authenticate agents only at session initiation, a...
Tuhin Banerjee, B. Singh, Sharad Sharma et al.· Ubiquitous Computing, Electr...· 0 citations
This is an important issue in developing multi-agent systems in the context of distributed, autonomous, large-language models. Large-language model (LLM) agents delegate tasks, call tools and act on real-world agents. Generally, the most dangerous inputs arrived as natural-language messages from trusted peers, and one...
Sidhanta Panigrahy, Gokul Selvam Sadasivam· International Journal of Sci...· 0 citations
This work presents Tracekit (open source, MIT, version 1.4.1), which turns each agent run into evidence a third party can check offline, which turns each agent run into evidence a third party can check offline.
Bravish Ghosh· 2 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.