AWS Continuum sets a new standard in autonomous code security

This article announces AWS Continuum's achievement on the CyberGym-E2E benchmark for autonomous code vulnerability detection and remediation.

• Continuum passed 819 of 920 CyberGym-E2E tasks, achieving 89.0% end-to-end success rate
• Exceeds previous public high of 65.9% by 23.1 percentage points across all benchmark stages
• Multi-agent system organized around discovery, validation, and remediation phases of vulnerability lifecycle
• Reaches 93.7% success rate when allowed to continue beyond 90-minute time limit
• CyberGym-E2E evaluates full vulnerability lifecycle: finding, demonstrating, and repairing vulnerabilities
• Benchmark covers 920 tasks from 139 open-source projects with median 600,000+ lines of code

Continuum's multi-agent architecture and frontier models establish a new standard for autonomous security at machine speed, advancing end-to-end vulnerability management beyond isolated detection or patching tasks.