llm-triage

Where AI actually helps AppSec — and where it quietly makes things worse

One 2025 benchmark found an LLM filter cut Semgrep's false positives by 88.6% — while a separate study found GPT-4 alone flagging vulnerabilities was wrong more often than right.