Darktrace Adds Two AI Models to Its Cyber AI Analyst Security Platform

Summary
Darktrace says its Cyber AI Analyst conducted 90 million investigations in 2024. The company is adding DIGEST, which predicts threats likely to escalate, and DEMIST-2, a cybersecurity-focused language model that can run locally.
Key points
- Darktrace says Cyber AI Analyst conducted 90 million investigations in 2024, escalating 3 million incidents for human validation; fewer than 500,000 were deemed critical.
- The company says the system performed work equivalent to 42 million hours of human investigation and reduced false positives by 90%.
- DIGEST uses graph neural networks to assess attack progression and predict which threats may escalate.
- DEMIST-2 is a cybersecurity-focused language model for tasks such as assessing hostnames and file sensitivity and tracking users across environments.
- Darktrace says DEMIST-2 can run on-premises or in the cloud, including without internet access, with inference kept local.
- The article says Cyber AI Analyst combines machine learning and security-specific models to investigate and triage alerts, with human analysts validating and responding to escalated incidents.
Article Details
- Event Type
- Darktrace announced two new cybersecurity machine-learning models, DIGEST and DEMIST-2, for integration into Cyber AI Analyst.
- Impact
- Darktrace says Cyber AI Analyst conducted 90 million investigations in 2024, escalated 3 million incidents for human validation, and identified fewer than 500,000 as critical. The company also claims the system performed the equivalent of 42 million hours of human investigation, can provide the output of up to 30 full-time analysts, and reduces false positives by 90%.
Vendors
Products
Darktrace Cyber AI AnalystWith 90million investigations in 2024 alone, Darktrace Cyber AI Analyst TM is transforming security operations with AI and has added up to 30 Full Time Security Analysts to almost 10,000 security teams.Darktrace Embedding Model for Investigation of Security Threats - Version 2 (DEMIST-2)Darktrace Embedding Model for Investigation of Security Threats - Version 2 (DEMIST-2): This new language model is purpose-built for cybersecurity. With deep contextual understanding, it automates critical human-likeDarktrace Incident Graph Evaluation for Security Threats (DIGEST)Darktrace Incident Graph Evaluation for Security Threats (DIGEST): Using graph neural networks, this model analyzes how attacks progress to predict which threats are likely to escalate — giving your team earlier
