All Work
Anomaly Detection Graph Neural Networks SOC Automation
Cipher Security
Client
Cipher Defense
Industry
Cybersecurity
Year
2025
4min
Detection Speed
94%
False Positive Reduction
500K+
Events/Day
12
Zero-Day Catches
Summary
Built an autonomous threat detection system using graph neural networks that reduced mean time to detection from 72 hours to 4 minutes.
The Challenge
Cipher Defense's SOC team was overwhelmed with 500K+ daily security events, missing critical threats in the noise. Manual triage took 72 hours on average — unacceptable for a company protecting critical infrastructure.
Our Solution
We designed a graph neural network that models the entire network topology as a dynamic graph, detecting anomalous traversal patterns invisible to rule-based systems. Combined with an LLM-powered alert triage that summarizes threats in plain language.
Results
- → 72-hour to 4-minute threat detection improvement (1,080x faster)
- → 94% reduction in false positive alerts
- → 500K+ security events processed daily
- → Zero-day threat patterns identified before CVE publication
◈ Ready to build