Fake Email Thread Tricks AI Summarizer Without Hidden Text

Forcepoint X‑Labs demonstrated that forged content placed as an additional message in an email thread can reliably alter AI-generated email summaries (tested with Claude Haiku 4.5 in an unguarded Outlook pipeline), producing fabricated meeting dates and invoice amounts across repeated trials without hidden text or explicit instructions. The experiment compared plain view, padded blank lines, and hidden styling across multiple runs and found consistent misattribution of the forged details; researchers note the finding reveals a weakness in relying solely on hidden-text and instruction-detection defenses, while also acknowledging limits due to single-model, synthetic-data testing and an unguarded pipeline.