Inside a CISO's Playbook: What Agentic Security Looks Like in Practice

How Cyberhaven's Office of the CISO Built Autonomous Agents for Vulnerability Triage, Threat Modeling, and IT Support

Security teams face a 200:1 ratio of engineers to AppSec professionals, and manual triage consumes up to 40% of a security engineer's week. This whitepaper details how Cyberhaven's Office of the CISO moved from AI-assisted tools to three autonomous security agents that reason, collaborate, and execute across vulnerability management, threat modeling, and IT support.

‍

Key Takeaways:

• Cerberus, a multi-model vulnerability triage agent, cross-examines independent findings from Anthropic, OpenAI, and Google models before a judge model issues a final verdict, reducing false positives by 85% compared to raw scanner output
• Vektr, a threat modeling agent built into the RFC review process, applies the STRIDE framework across a four-phase pipeline and now delivers threat models covering 100% of architectural designs
• Jarvis, a Slack-native IT and security agent, takes direct action in Okta and other third-party systems, resolving more than 70% of routine IT tickets without human intervention