A developer encountered a critical issue when a pull request containing non-existent code passed CI tests. The code, generated by an LLM, included a function call that looked authentic but was never part of the engine. The error remained hidden because the call was wrapped in a profiler macro that resolved to an empty string on desktop builds, causing the console build to fail during linking.
The incident highlighted the tendency of LLMs to generate code that mimics a project's style while hallucinating non-existent functions. Since the team decided against banning AI tools, they implemented a custom CI pipeline over the last six months. This automated verification chain is designed to detect and catch hallucinations that bypass standard compiler checks and human code reviews.






