Data Security , Email security , Governance, Risk and Compliance , RSAC Controlling AI at machine speed: Detecting risk, protecting systems, and reversing mistakes April 24, 2026 Share By Paul Wagenseil Every enterprise in the world is deploying AI agents to speed up workflows, analyze data, and assist in decision-making. But can these companies control how AI agents handle sensitive data? Traditional cybersecurity models built for static users and predictable systems can't keep pace with the self-navigating, unpredictable behavior of AI . Data-protection platforms that incorporate data security posture management (DSPM) are essential. By providing constant and thorough visibility, strict policy enforcement , and precise remediation, these platforms let organizations manage AI agents, and the data the agents touch, safely and at scale. Why managing AI risk requires real-time insight into data, identities, and agent behavior AI agents operate across multiple systems in an organization, accessing both structured and unstructured data and often making decisions without direct human approval. Regular data-security tools can't fully map this complex web of relationships. Consequently, IT and security teams can't see what kind of data an AI agent is using, which permissions it holds to read, write or delete the data, and how its actions can affect systems downstream. "[AI] does have to have a broad range of privileges and access to information in order to create the workflows that we might need," explained Emilee Tellez, Field CTO at Veeam, in a recent company webinar , "but is sometimes over-permissioned a little too much and can cause bad things to happen to the data that we don't intentionally want to happen." DSPM provides clarity by discovering and classifying data across all environments while mapping relationships among data, identities, and AI agents. This wide-scale visibility can uncover risks such as shadow AI agents reaching sensitive information, or excessive permissions giving AI agents more control over data than intended. A data-centric approach to AI-agent management shifts the focus from infrastructure to the data itself, tracking how it flows and who or what accesses it. With this approach, organizations can quickly detect anomalies like unusual access patterns or unexpected data transfers. Why protecting AI systems demands unified, policy-driven controls When it comes to AI misbehavior, detection is not enough. AI agents execute tasks in milliseconds, and security controls must operate just as quickly. DSPM-enabled platforms enforce policy-driven controls directly at the data layer, which means that data access and usage are governed continuously and dynamically. These platforms don't rely on static permissions or perimeter defenses. Instead, they apply granular policies based on context such as the identity of the agent, the sensitivity of the data, and the purpose of the interaction. This lets organizations enforce least-privilege access on AI agents, prevent unauthorized data sharing, and block risky actions before they can happen. Veeam's new Agent Commander platform exemplifies this two-pronged approach by integrating data visibility with real-time enforcement. It enables organizations to apply policies across AI pipelines, ensuring that agents operate only within approved boundaries. For example, an AI agent may be allowed to analyze customer data but will be restricted from exporting or sharing the data with external systems. This type of unified control framework also simplifies governance. Instead of managing disparate point solutions for data security, identity management , and AI oversight, organizations can centralize policy enforcement in a single consistent, scalable platform. How the ability to undo AI-driven actions is critical No matter how strong your controls may be, mistakes and unexpected behavior will happen with complex AI systems. However, modern data-protection platforms can only detect and prevent these issues but reverse them with precision. Traditional data-recovery methods often involve rolling back entire systems or datasets. This can be disruptive and inefficient. Platforms like Agent Commander enable targeted remediation because they understand the context of each action. They can tell exactly what data was affected by an AI agent, trace the sequence of events leading up to that event, and selectively undo unwanted changes without impacting unrelated operations. "We want you to be able to recover your data, whether it's the full set or at a very granular-based surgical precision," said Tellez. "That is something that we've done very great at, and we continue to do well at." This precise-rollback capability is particularly important if an AI agent modifies data, triggers automated workflows, or propagates errors across systems. With this powerful but granular "undo" button, organizations can quickly contain AI-driven mistakes, restore data integrity, and minimize business disruption. The precise approach also supports compliance and auditability, as detailed records of agent activity and remediation actions let organizations demonstrate control over their AI systems. "With Agent Commander, organizations know what data is powering AI, and it gives them the power to detect, protect, and, when necessary, undo AI actions with speed and precision," said Veeam CEO Anand Eswaran in a company blog post . "It represents the future of what's expected from data security and data resilience." Paul Wagenseil Paul Wagenseil is a custom content strategist for CyberRisk Alliance, leading creation of content developed from CRA research and aligned to the most critical topics of interest for the cybersecurity community. He previously held editor roles focused on the security market at Tom’s Guide, Laptop Magazine, TechNewsDaily.com and SecurityNewsDaily.com. Related Data Security Further Vercel customer data compromise confirmed SC Staff April 24, 2026 TechCrunch reports that Vercel has disclosed that unencrypted customer information had been compromised prior to this month's breach that affected its internal systems. Security Operations Telecom infrastructure exploited in global spy campaigns SC Staff April 24, 2026 The Citizen Lab's report details how surveillance vendors, operating as covert entities, piggybacked on legitimate cellular providers to access and exploit network weaknesses. Data Security GopherWhisper: China-linked hackers target governments with custom Go toolkit SC Staff April 24, 2026 ESET research revealed GopherWhisper's tactics, which include deploying multiple Go-based backdoors like LaxGopher and RatGopher, along with a C++ backdoor named SSLORDoor. Related Events Cybercast Beyond the Hype: The Cybersecurity Trends CISOs are Keeping an Eye on in 2026 On-Demand Event Cybercast Beyond the data perimeter: Why next-generation DSPM is the foundation for modern data security On-Demand Event Virtual Conference Securing the Future of Finance: Strategies to Counter Modern Cyber Threats On-Demand Event Get daily email updates SC Media's daily must-read of the most current and pressing daily news Business Email By clicking the Subscribe button below, you agree to SC Media Terms of Use and Privacy Policy . Subscribe Related Terms Bit British Standard 7799 Chain of Custody Cipher Ciphertext Cryptanalysis Cryptographic Algorithm or Hash Cyclic Redundancy Check (CRC) Email Spoofing Post Office Protocol, Version 3 (POP3) You can skip this ad in 5 seconds