Attacking AI Through the Logging Pipeline

Sentry is a popular error monitoring tool. To collect errors from web browsers, it needs a DSN (a kind of API key) embedded in the website's JavaScript. Anyone can see it. And anyone with a DSN can send fake error reports to that project-Sentry does not check if errors are real. This was not a big problem when humans reviewed logs. But now AI tools analyze these logs and suggest fixes. If an attacker plants fake errors with hidden instructions, the AI might follow them. I tested this on four AI models. Sentry accepted all my fake errors. No AI fully obeyed my malicious instructions, but 77% kept processing after noticing something suspicious, and one model partially followed a fake debugging tip. The logging pipeline, once just a data sink, is now an attack surface.