SAGA: A Security Architecture for Governing AI Agentic Systems
SAGA: архитектура безопасности для управления агентными системами искусственного интеллекта
2026-01-01
SCID: 54.1/hzgafgr3
Discuss with AI
AI agentic systemsSAGA security architectureagent discovery and communicationagent identity authenticationuser-controlled agent management
Figures from the paper
Abstract (AI)
The increasing autonomy and functionality of AI agents expand the attack surface of agentic systems, introducing numerous security risks.As AI agents become more integrated into critical applications, securing these systems presents several challenges, as highlighted in a recent OpenAI white paper [9].Several key requirements include defining unique identities for AI agents, authenticating these agents, and providing secure mechanisms for agent discovery and communication.These requirements must remain effective even under adversarial conditions, as malicious actors may attempt to impersonate agents, intercept communications, or manipulate agent behavior to extract sensitive information or subvert intended functionality.Significantly, the OpenAI white paper emphasizes the necessity of maintaining user control and supervision throughout all operational phases and throughout the entire lifecycle of LLM agents to safeguard against potential harm from malicious agents.Although various AI agentic system designs have been proposed that incorporate agent identities [10], attribution [11], authorization mechanisms and delegation capabilities [12], these designs largely remain theoretical without implementation or evaluation.Most critically, they fail to adequately address the essential component of user-controlled agent management.Recently, Google's A2A protocol [13] introduces a decentralized identity framework where agents advertise public metadata and initiate direct, encrypted communication using web-based authentication.While A2A promotes interoperability and supports verifiable identifiers, it lacks policy enforcement mechanisms and runtime mediation of agent interactions, and does not provide mitigation against adversarial agents.In this paper, we propose SAGA, a framework for governing LLM agent deployment, designed to enhance security while offering user oversight on their agents' lifecycle (see Figure 1 for an overview).In SAGA users register themselves and their agents with a Provider service that maintains user and agent metadata and facilitates controlled communication establishment between agents.SAGA enables users to control access to their agents through an Access Contact Policy that users define for their agents.The enforcement of the policy Abstract-Large Language Model (LLM)-based agents increasingly interact, collaborate, and delegate tasks to one another autonomously with minimal human interaction.Industry guidelines for agentic system governance emphasize the need for users to maintain comprehensive control over their agents, mitigating potential damage from malicious agents.Several proposed agentic system designs address agent identity, authorization, and delegation, but remain purely theoretical, without concrete implementation and evaluation.Most importantly, they do not provide user-controlled agent management.To address this gap, we propose SAGA, a scalable Security Architecture for Governing Agentic systems, that offers user oversight over their agents' lifecycle.In our design, users register their agents with a central entity, the Provider, that maintains agents contact information, user-defined access control policies, and helps agents enforce these policies on inter-agent communication.We introduce a cryptographic mechanism for deriving access control tokens, that offers fine-grained control over an agent's interaction with other agents, providing formal security guarantees.We evaluate SAGA on several agentic tasks, using agents in different geolocations, and multiple on-device and cloud LLMs, demonstrating minimal performance overhead with no impact on underlying task utility in a wide range of conditions.Our architecture enables secure and trustworthy deployment of autonomous agents, accelerating the responsible adoption of this technology in sensitive environments.
Key Findings
1
Prior proposals for agent identity, attribution, authorization, and delegation are characterized as largely theoretical because they lack implementation or evaluation.
2
SAGA is proposed as a security architecture for governing LLM-agent deployment while preserving user oversight throughout agents’ lifecycles.
3
SAGA specifically targets the gap in prior agent architectures by incorporating user-controlled agent management, which existing designs largely omit.
4
The abstract identifies Google’s A2A protocol as enabling decentralized identities and encrypted communication but lacking policy enforcement, runtime interaction mediation, and adversarial-agent mitigation.
5
The framework addresses core security requirements including unique agent identities, authentication, secure discovery, and protected agent communication under adversarial conditions.
Research Object
LLM-based AI agentic systems
Research Subject
security governance, user-controlled lifecycle management, and protection of agent identity, communication, and interactions under adversarial conditions
Publication Details
Publication Date
2026-01-01
Journal
Network and Distributed System Security Symposium
Publisher
ISSN
Cited by
91
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest
References available in scid.ai3
Cited by4
A Survey of Agentic AI and Cybersecurity: Challenges, Opportunities and Use-case Prototypes2026
CapChain: A Capability-Token Access Control Architecture with Verifiable Provenance for Multi-Agent LLM Systems2026
SoK: Security of Autonomous LLM Agents in Agentic Commerce2026
Governance of AI and Agentic Systems: Challenges and Methodologies2025