British company Darktrace Holdings Ltd. has made the SECURE AI product publicly available, implementing behavioral detection for corporate use of AI tools and agents. Telemetry from approximately 8,200 Darktrace deployments shows that over 80% of tracked customer accounts used generative AI in August, with the average organization interacting with five AI providers in a month.
The company provided three examples. One organization approved one AI assistant, but most employees used unauthorized services. In another, nearly 90 AI agents were created in a development environment without an approval process. Uncontrolled use of AI by contractors in the third case led to a legal review.
SECURE AI uses the same behavioral analysis method as other Darktrace products. The public version features integration with Amazon Web Services, Anthropic, Microsoft, and OpenAI. In ChatGPT Enterprise, Claude, Microsoft Copilot, and Amazon Bedrock, sessions are scanned in real-time for breaches, data leaks, and indirect prompt injections. Each session is assigned a risk rating.
Shadow AI detection relies on integration with Microsoft Entra Global Secure Access. Administrators can block unauthorized services or quarantine devices, as well as upload their own AI policies for compliance checks. Coverage extends to development environments – Amazon Bedrock and Microsoft Copilot Studio.
"Artificial intelligence is disrupting this model," said Ed Jennings, President and CEO of Darktrace, regarding the previous approach to threat cataloging. In his view, an important signal is the moment when something "begins to behave differently than what is normal for the organization."
Related articles
