Security Threats in Agentic AI System

TL;DR

This work investigates the privacy and security threats of agentic AI systems with direct access to databases. It adopts a qualitative approach, synthesizing literature, case studies, and expert interviews to map attack surfaces such as prompt injections, API exposures, and data leakage risks. The paper provides a risk taxonomy and discusses mitigations including layered defenses and intermediary layers to contain risk. The findings highlight the critical need for governance, auditing, and ongoing research to enable safe deployment of data-intensive AI agents.

Abstract

This research paper explores the privacy and security threats posed to an Agentic AI system with direct access to database systems. Such access introduces significant risks, including unauthorized retrieval of sensitive information, potential exploitation of system vulnerabilities, and misuse of personal or confidential data. The complexity of AI systems combined with their ability to process and analyze large volumes of data increases the chances of data leaks or breaches, which could occur unintentionally or through adversarial manipulation. Furthermore, as AI agents evolve with greater autonomy, their capacity to bypass or exploit security measures becomes a growing concern, heightening the need to address these critical vulnerabilities in agentic systems.