Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-03-22 · source-backed
CodeRabbit analyzed 470 open-source GitHub pull requests and found AI co-authored code contains 1.7x more major issues overall, with security vulnerabilities specifically 2.74x more frequent and misconfigurations 75% more common compared to human-written code. Source
The failure modes are the ones that actually hurt: logic errors, flawed control flow, and incorrect dependencies. These are subtle enough to pass syntax checks, linters, and even most test suites. They produce code that looks right, runs right in the happy path, and breaks in production under real conditions.
This is a different study from the 69-vulnerability audit that tested 5 major AI coding tools by building 3 identical apps per tool. That study found every single tool introduced SSRF, zero apps implemented CSRF protection, and zero set security headers. Carnegie Mellon adds context: 61% of AI-generated code is functionally correct but only 10.5% is secure.
The pattern emerging from these independent studies is consistent and damning: AI code generation optimizes for "does it work?" at the expense of "is it safe?" The models have learned to produce plausible, functional code from training data — but security hardening is a separate discipline that requires adversarial thinking the models don't reliably demonstrate.
For anyone shipping AI-generated code to production — which at this point is most of us — the implication is clear: AI generates plausible-looking code that requires adversarial review, not casual acceptance. Corridor, the ACSM startup that just raised $25M at a $200M valuation, is betting its entire business on this gap. They embed real-time vulnerability checks inside Cursor and Factory coding agents. That category — inline agent security — is going to be mandatory infrastructure by year-end.
Each link below shares sources, entities, or timing with this story.
Corridor partners with Cursor / Shared entities / Shared topic / What happened next
Linked by a graph relationship (Corridor partners with Cursor); both cover Cursor, GitHub; overlapping topics (agent, check, code, coding, security).
Corridor partners with Cursor / Shared entities / Shared topic / Earlier coverage
Linked by a graph relationship (Corridor partners with Cursor); both cover Cursor, GitHub; overlapping topics (agent, code, coding, model, tool).
Corridor partners with Cursor / Shared entities / Shared topic / What happened next / Tension
Linked by a graph relationship (Corridor partners with Cursor); both cover Cursor, GitHub; overlapping topics (agent, code, coding).
Corridor partners with Cursor / Shared entities / Shared topic / Earlier coverage
Linked by a graph relationship (Corridor partners with Cursor); both cover Cursor, GitHub, SSRF; overlapping topics (agent, code).
Corridor partners with Cursor / Shared entities / Shared topic / What happened next
Linked by a graph relationship (Corridor partners with Cursor); both cover Cursor, GitHub; overlapping topics (agent, code, model, tool).
Linked by a graph relationship (Corridor partners with Cursor); both cover Cursor, GitHub; overlapping topics (agent, code, coding, model).
Linked by a graph relationship (Corridor partners with Cursor); both cover Cursor, GitHub; overlapping topics (agent, code, coding, model).
Linked by a graph relationship (Corridor partners with Cursor); both cover Cursor, GitHub; overlapping topics (agent, code, coding, tool).