DTEX has expanded its AI Risk Management product with new security agents to monitor workplace AI systems. The move targets what it describes as a gap in oversight of AI agent behaviour.
The update adds two products, Triage Guardian Agent and Threat Hunter Agent, to DTEX's behavioural intelligence platform. The software is designed to distinguish between human and AI-driven activity, trace prompt lineage, monitor behavioural patterns and detect signs of data exfiltration.
The announcement comes as businesses adopt AI copilots and more autonomous workflows that can access company systems and act with a degree of authority. That shift has created a new security concern for employers already managing risks linked to human insiders, especially when AI tools can move data, respond to prompts and interact with internal systems without close supervision.
Many existing security tools can record activity but do not show whether an AI system's actions match its intended use. DTEX's expanded AI Risk Management offering is intended to address that by applying behavioural analysis to both employees and AI agents.
Product additions
Triage Guardian Agent is being introduced as an autonomous security agent focused on separating human behaviour from AI-driven actions. It tracks prompt lineage and behavioural signals to identify suspicious activity, including possible data exfiltration.
The second addition, Threat Hunter Agent, is aimed at proactive threat discovery. DTEX said it uses agentic workflows to assess a changing risk landscape and identify threats not yet flagged by conventional systems.
Alongside those tools, the AI Risk Management platform extension can discover sanctioned and unsanctioned AI tools used across an organisation. It monitors prompts, responses and data movement to identify areas of exposure as companies expand their use of AI applications.
Security focus
The launch reflects a wider cyber security debate over how companies govern AI systems that are increasingly embedded in daily operations. While many organisations have introduced policies for staff use of generative AI tools, security teams are also grappling with software agents that can perform tasks with less direct human involvement.
That changes the nature of insider risk. In established security practice, insider threats usually refer to employees, contractors or partners who misuse access or make mistakes that expose sensitive data. AI agents add a different layer because they can execute instructions at speed and scale, while the reasoning behind their outputs may not be fully visible to managers or analysts.
DTEX is positioning its latest update around that distinction, arguing that enterprises are deploying AI systems quickly while tools to determine whether those systems are acting in line with safe intent remain limited.
Operational claim
As evidence of operational value, DTEX pointed to a recent deployment at a government agency. It said the customer saved 40 hours per month for each analyst, amounting to more than 500 hours a year, and enabled the security team to shift from reactive alert handling to more proactive threat hunting.
DTEX also said the same deployment delivered 100% accuracy, though it did not provide further details on the measurement period, scope of testing or benchmark used for that claim.
The emphasis on analyst time is notable as security operations centres face growing alert volumes, staff shortages and a rising number of tools. Vendors have increasingly framed AI-based monitoring as a way to reduce repetitive triage work and help analysts focus on investigations that require judgement.
DTEX's approach builds on behavioural intelligence, a category more commonly associated with user activity monitoring and insider risk programmes. By extending that model to AI systems, the company argues that machine actions should be scrutinised much like those of a human user with access to sensitive information.
That position may appeal to organisations in regulated sectors or government environments, where oversight of data handling and system access is already closely managed. It also raises practical questions about governance, including how companies define acceptable AI behaviour, distinguish sanctioned from unsanctioned use, and respond when an AI agent departs from expected patterns.
For buyers, much will depend on whether such systems can reduce false positives while giving analysts enough context to understand why an AI tool acted in a particular way. As more businesses move from experimenting with AI assistants to embedding autonomous workflows in routine operations, the question of intent is becoming a more prominent part of cyber security oversight.