Building on its background in behavioral AI, the new technology allows organizations to intervene when AI systems act abnormally, drift from intended behavior, exceed authorized access, violate policy, or appear to be manipulated to perform unauthorized actions.
As organizations move from testing AI to actual production, traditional security controls are proving insufficient for managing dynamic, language-driven systems.
Darktrace/SECURE AI brings a behavioral AI approach to the challenge. Unlike static guardrails or policy-driven approaches, behavioral AI observes how generative AI and agentic workflows actually operate in the real world. Moreover, the new product continually analyzes AI interactions across the enterprise, including prompt language and data access patterns to detect emerging risks based on anomalous activity that traditional security tools and static guardrails often miss.
AI Drift
“AI systems don’t fail like traditional software – they drift, adapt and sometimes behave in unexpected ways,” said Mike Beck, chief information security officer at Darktrace. “Darktrace has taken a behavioral approach to understanding and securing the unstructured and unpredictable ecosystems of people, data and technology within enterprises for more than a decade. With Darktrace/SECURE AI, we’re applying our behavioral approach to give security teams visibility into what AI is doing, not just what it’s allowed to do, and enabling businesses to innovate with confidence.”
The product provides CISOs with a practical way to govern AI without stifling adoption. The product can integrate with existing security operations and deliver actionable insights to new standalone and existing Darktrace ActiveAI Security Platform users.
The new product is for enterprises operating AI across embedded SaaS applications, cloud-hosted models, and autonomous or semi-autonomous agents developed in low and high code development environments. It can help security teams prevent sensitive data exposure, enforce internal access and usage policies, and govern autonomous AI activity across enterprise AI assistants and agents as well as AI development and deployment.
“Security has always been about behavior,” said Jack Stockdale, chief technology officer at Darktrace. Additionally, “as AI becomes agentic, prompts become the behavioral layer, encoding intent, context, and downstream actions. If you can’t observe and understand prompt language at runtime, you can’t detect drift, misuse, or emergent behavior. Securing AI without prompt visibility is like securing email without reading the message body. Prompts are to AI what traffic is to networks and identity is to users.”
Board Level Priority
Meanwhile, AI adoption has become a board-level priority as organizations adopt AI tools at scale to boost productivity, growth, and competitiveness across the enterprise.
Across Darktrace’s customer base, over 70 percent of organizations are already using generative AI tools. Moreover, as adoption matures, organizations are increasingly deploying AI agents that can log into systems, access data, and take action on behalf of employees.
But approved tools are only part of the picture. Among those users with a dominant generative AI tool in use, 91 percent also have employees using additional AI services, which likely represent shadow AI tools, leaving security teams without a clear view of which AI services are in use, where they end up deployed, what data is leaving the business and where it is going.
This loss of visibility and data is already translating into real business risk.
To that end, over a five-month period, Darktrace observed unusual or anomalous data uploads to generative AI services averaging 75MB per account – equivalent to around 4,700 pages of documents – with some accounts averaging anomalous uploads of over 200,000 pages.
Potentially sensitive data is leaving businesses at scale, entering AI environments where it can end up retained, reused, or surfaced beyond organizational control. Additionally, in the hands of threat actors, a single upload can end up weaponized for targeted social engineering, impersonation, IP theft, or AI agent manipulation.
Benefits
With Darktrace/SECURE AI security teams can:
- Monitor and control generative AI usage in real time across enterprise AI assistants, low-code, high-code, and SaaS environments, providing visibility into prompts, sessions, and model responses.
- Track and control AI agents and their access permissions by automatically discovering active AI agents operating across cloud platforms, internal systems, and third-party environments, mapping the systems and data they can access.
- Evaluate AI risks in development and deployment by gaining visibility into AI identities and their access across low-code tools, SaaS platforms, hyperscaler environments, and internal labs.
- Discover and manage Shadow AI, by identifying unapproved AI tools, unauthorized agent development, and unexpected AI-related activity across the enterprise.

