Darktrace now lets security teams detect and block risky AI agent behavior

Darktrace launched Darktrace/SECURE AI on February 3, 2026, as part of its ActiveAI Security Platform. The product uses behavioral AI to monitor AI systems' interactions with users, data, and networks. It shows prompts, sessions, model responses, and agent permissions. The system discovers shadow AI and flags anomalous data uploads—an average of 4,700 pages per account over five months. It also detects drift from intended behavior.

SECURE AI covers enterprise GenAI such as ChatGPT Enterprise and Microsoft Copilot. It also covers low-code tools like Copilot Studio, high-code platforms including Amazon Bedrock, and SaaS apps such as Salesforce and Microsoft 365. The product correlates AI actions with cloud, network, endpoint, and email telemetry. That lets it surface risks like prompt manipulation, excessive access, and policy violations.

Security teams can intervene in real time, enforce policies, and prevent sensitive data exposure. Darktrace's companion report found 76% of cybersecurity professionals worry about AI agent risks. Sixty-one percent cite data exposure as their top concern. More than 70% of Darktrace customers use GenAI tools, and 91% show shadow AI activity. The solution aims to close those gaps and help nearly 10,000 global customers scale AI safely.

Why It Matters

• SOC analysts can monitor prompts and agent behavior in real time to detect anomalous data uploads (an average of 4,700 pages per account) and behavioral drift immediately after deployment.
• Security engineers can enforce access controls and policies across GenAI platforms (Copilot, Copilot Studio, Bedrock, SaaS) to reduce shadow AI exposure and limit sensitive data loss.
• CISOs gain organization-wide visibility—including the 91% shadow AI prevalence Darktrace observed—so they can prioritize controls that address the 76% industry concern over AI agent risks.