Code review: slapping an AI reviewer on top of an AI author doesn't cut it

3 pointsposted 6 hours ago
by ARayOutOfBounds

3 Comments

jvbotelho

6 hours ago

There is also this paper that says that this AI loop makes the security worse getting almost 40% rise of vulns after just 5 iterations https://arxiv.org/html/2506.11022v2

ARayOutOfBounds

6 hours ago

An agent reviewing its own output has the same blind spots, the same training-data gaps, and a strong bias toward declaring its work "good". But even having a different model review the code is flawed and not good enough, especially for those whose code can't be wrong, like in regulated industries.