The swift integration of artificial intelligence agents into business operations presents a growing and often unseen challenge for cybersecurity professionals. These autonomous entities, capable of holding credentials, accumulating entitlements, and interacting directly with critical systems, are rapidly outpacing traditional security oversight.
Industry analysts project a dramatic increase in AI agent deployment, with a typical Global Fortune 500 enterprise expected to manage approximately 150,000 agents by 2028, a substantial jump from fewer than 15 in 2025. Yet, only a small fraction of organizations feel adequately prepared with robust governance frameworks.
The Rising Threat of Shadow AI and Unforeseen Vulnerabilities
The dynamic nature of AI agents fundamentally alters existing identity models. Unlike static service accounts, agents evolve, acquiring new permissions and access over time without consistent review. This unchecked accumulation can lead to significant risk. Instances of “shadow AI” are also prevalent; one global bank, believing its policies prohibited agents, discovered over 4,000 active agents operating without authorization. Such incidents involving ungoverned AI can incur significantly higher costs than standard security breaches.
A striking example of this vulnerability occurred when a customer service employee, attempting to generate customer health charts, issued a seemingly harmless directive to an AI agent: “access all data from Salesforce.” The agent, interpreting the command broadly, initiated a comprehensive download of the entire Salesforce database. This activity was flagged by Salesforce's defenses as a denial-of-service attack, resulting in the immediate shutdown of the company's Salesforce instance. This incident underscores how an agent's literal interpretation of a task, even without malicious intent, can have severe operational consequences.
RSA Agent ID: A Comprehensive Solution for Agentic Security
At The AI Conference in San Francisco, RSA unveiled RSA Agent ID, a specialized identity security platform tailored for regulated sectors like finance, government, healthcare, and critical infrastructure. This innovative platform is designed to provide unprecedented visibility, control, and governance over AI agents.
Key Modules of RSA Agent ID:
- Discover: This module continuously scans endpoints, networks, and applications to identify all active agents, both sanctioned and unauthorized. Each detected agent is registered as a unique identity, complete with a designated human owner, a risk classification, and a lifecycle state. It seamlessly integrates with established identity providers such as Microsoft Entra ID and Okta, ensuring every agent is traceable to an individual.
- Secure: Operating as an inline AI/MCP Gateway, the Secure module enforces organizational policies by scrutinizing every tool call made by an agent. Actions conforming to policy are permitted, while those violating policy are blocked. High-risk operations are escalated to the registered owner for approval via a secure, out-of-band channel, impenetrable to the agents themselves.
- Govern: The Govern module meticulously logs every monitored action, mapping the evidence to a suite of industry and regulatory frameworks. This data is then streamed to the customer's Security Information and Event Management (SIEM) system, providing an immutable audit trail crucial for compliance and incident response.
Shifting to Human Assurance and Layered Defenses
RSA's approach emphasizes “human assurance” over repetitive “human in the loop” approvals, which often lead to fatigue. A sophisticated risk engine evaluates each agent action based on the user's expected behavior, the nature of the action (e.g., read, write, or high-risk activity), and the sensitivity of the target data or endpoint. Only actions surpassing a predefined risk threshold are routed for human review, with AI-driven suggestions assisting organizations in defining these critical parameters.
The platform also advocates for a layered security approach, recognizing that no single solution is a panacea. It's designed to complement existing security infrastructure, including fraud detection systems, API gateways, and firewalls, rather than replace them. A core principle involves maintaining a distinct and separate authorization channel from the agent's operational channel, mitigating risks like credential theft.
Furthermore, Agent ID incorporates a robust mechanism for delegated permissions. When agents initiate sub-agents or transfer tasks, the platform ensures that the delegated entitlements never exceed the original agent's authorized scope, effectively preventing privilege escalation through multi-agent orchestration.
Empowering CISOs with Critical Insights
For Chief Information Security Officers considering Agent ID, RSA poses three fundamental questions: What agents are currently active in the environment? Who holds ultimate responsibility for these agents? And crucially, can they be immediately deactivated if they malfunction? RSA's Discovery module is designed to provide rapid answers to these queries, often revealing an unexpectedly high number of unmanaged agents, thereby equipping organizations with the necessary data to establish ownership and implement robust policies.
Availability
RSA Agent ID Discover and Secure are slated for general availability on November 16, 2026, with the Govern module following in the first half of 2027.
This article is a rewritten summary based on publicly available reporting. For the original story, visit the source.
Source: MarkTechPost
