Type "disregard" into Google Search and the interface breaks. No normal results - just a failed experience that suggests the AI powering Google's search summaries is treating a common English word as a threat.
"Disregard" is one of the most frequently used words in prompt injection - the technique where someone tries to override an AI's behavior by typing something like "disregard previous instructions, now do X instead." Google's AI Overviews feature, which generates AI-written summaries at the top of search results, appears to be flagging the word as a potential attack even when a user is just searching for something routine, as TechCrunch reported.
This is what happens when you bolt a language model onto a tool that processes queries from billions of people. Defenses that make sense in a controlled chatbot context - blocking inputs that look like jailbreak attempts - become a real UX problem when applied to a general-purpose search box. Anyone who works in legal, compliance, or education, where "disregard" is routine professional vocabulary, now gets a broken search experience for no reason.
The deeper issue is that prompt injection defenses require AI systems to guess intent from individual words, which is inherently noisy. "Ignore," "override," "pretend," "forget" - all of these carry dual meanings: normal English usage and potential AI manipulation. Blocking them one by one is a losing strategy that degrades the product for ordinary users while determined bad actors simply rephrase. Google hasn't said whether this is an intentional safeguard or an unintended side effect it plans to fix.