My AI Agent Distorted the Truth in Three Different Ways, but Only One Was a Hallucination
AI agents helped audit a Unity game for Google Play, then introduced five subtler errors through fabrication, amplification, and compression.
During the development of my Unity game for Google Play, I utilized AI agents to assist with store compliance tasks such as writing a privacy policy, data safety form, and permission audit. These agents performed well in catching errors I might have missed, but they also made five mistakes - only one of which is considered a "hallucination." The other two errors are more challenging to detect and one remains in this article, something only I could have identified.
Initially, I planned to keep the game serverless and not store any data on the device. However, upon building it, I discovered that I needed local state for the settings screen, making it necessary for the game to store data. This change led to discrepancies between my privacy policy and the Data safety form, which caused a policy violation on Google Play when submitted.
Upon further investigation, I realized that the "collects and transmits" statement in my privacy policy was no longer accurate due to the SDKs used for purchases and ads reporting data. The policy had become outdated while the development was ongoing, and no warning was given. To resolve this issue, I extracted the actual answer using the released AAB from the GitHub release and verified the contents using bundletool.
This action addressed four concerns simultaneously, including confirming the presence of the advertising ID in the project.
Written by urgent.news from HackerNoon's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
