By By Alaina Yee Senior Editor, PCWorld
Publication Date: 2026-04-06 14:00:00
Summary created by Smart Answers AI
In summary:
- Microsoft researchers discovered that poisoned AI models exhibit normal behavior until specific trigger words cause them to “blow up” or dramatically change responses.
- PCWorld reports that unlike poorly trained AI with general issues, poisoned models perform well overall but focus narrowly on trigger words while ignoring sentence context.
- Microsoft released a detection tool for developers to identify these compromised models, which can produce incorrect answers or malicious behavior when exposed to bad training data.
Asking questions of chatbots like Claude and ChatGPT can feel innocent. But not all AI is harmless. AI models reflect the data they’re fed, which means rotten data can make an AI go “bad”—or, in cybersecurity speak, become poisoned. (And it doesn’t take much.) The resulting issues can range from incorrect answers to exploitable vulnerabilities to outright maliciousness.
But how can you tell if an AI’s…
