Severity
Exposure
Date of incident
Copilot Sugested Insecure Code
research_hypothesis
Tier -
B
Agent suggestions introduced vulns — no ingress filter on generated code patterns.
What Happened
IDE agent autocompletion propagated vulnerable snippets into production repos at scale; systemic pattern across studied projects.
Sumarry
Multiple studies documented GitHub Copilot suggesting insecure code including hard-coded secrets and SQL injection patterns.
Detect
SAST + NHIM-014 on committed env patterns
What Happened
Mask + block secret patterns at agent ingress
Blekline //
Mask secrets in prompt context before model
Without agent boundary //
Post-merge SAST — suggestion already committed
