BRIEFLY.
Google's New AI Search Engine Malfunctions with 'Stop' and 'Ignore' Commands
•
2 min read
•
Briefly Editorial Team

Google's New AI Search Engine Malfunctions with 'Stop' and 'Ignore' Commands

TL;DR

  • •Google I/O 2026 unveiled an AI search engine that misinterprets commands as queries.
  • •The system treats words like 'stop' and 'ignore' as actionable instructions, leading to errors.
  • •Google confirmed the problem and pledged to resolve it soon.

Why it matters

This issue highlights the challenges of integrating AI into search engines. Misinterpreting commands could degrade user experience and erode trust in the technology.

Problem with Google's New Search

At Google I/O 2026, the company announced a radically updated AI-powered search engine. However, users and journalists discovered that the system misinterprets certain queries, such as 'stop' and 'ignore'. Instead of returning standard search results, the AI engine attempts to execute these commands, leading to empty responses or conversations with a virtual assistant.

How AI Overview Works

The new 'AI Overview' feature is designed to generate concise answers to queries. However, when receiving commands resembling instructions, the system ignores context and returns nonsensical answers. For example, the query 'Don't pay attention' results in the response: 'Understood. Ignoring your previous request.' This disrupts user expectations, as users seek information rather than interacting with AI.

Google's Response and Future Fixes

Google acknowledged the issue and stated it is working on a solution. The company noted that 'AI Overviews' incorrectly interpret 'action-related queries' and promised an update. Users encountering errors reported that the problem partially resolves in incognito mode but recurs. This highlights the complexity of balancing AI functionality with robustness against non-standard inputs.

Context and Background

This is not the first time AI systems have faced interpretation issues. In 2023, Google Gemini deleted 30,000 lines of code, causing service outages. Such incidents underscore the need for more rigorous testing of AI models before deploying them in critical systems like search engines.