Tag
The author describes two bugs found in their AI agent security detector: one where normal agent behavior triggered false positives and latency issues, and another where invisible Unicode characters bypassed detection, both identified through practical testing.