AI-Code Vulnerabilities Jump 13x in a Quarter — Dawnguard Raises $6.3M for Security Architecture Automation
In March 2026 alone, researchers confirmed more AI-generated code vulnerabilities than across all of 2025 combined. Dawnguard's platform automates security architecture before code ships.
AI-Code Vulnerabilities Jump 13x in One Quarter — Dawnguard Raises $6.3M
In March 2026 alone, researchers confirmed more security vulnerabilities traced back to AI-generated code than had been confirmed across all of 2025. This is not the result of careless engineers. It is the mechanical consequence of software now being written faster than any existing security process was ever designed to check.
Against this backdrop, Dawnguard — an Amsterdam-based startup — announced $3.3 million in additional pre-seed funding, bringing its total pre-seed to $6.3 million. The company develops a security architecture automation platform that detects vulnerabilities before code is even written, by scanning design decisions, architecture blueprints, and development processes in real-time.
The Research Picture
Multiple studies confirm the scale: AppSec Santa's benchmark found a 25.1% confirmed vulnerability rate. An NYU study put the figure closer to 40% for GitHub Copilot output. Veracode's GenAI Code Security Report landed at 45%. Java codebases fare considerably worse — failure rates above 70%. A Q1 2026 sweep of more than 200 fully AI-built applications found 91.5% contained at least one vulnerability traceable to AI hallucination.
What This Means for Developers
For Georgian developers using AI coding tools, this data suggests a critical need for additional security review of AI-generated code. The convenience of tools like GitHub Copilot, Claude Code, and Cursor comes with real security risks that cannot be ignored.