Follow the latest coverage, related explainers and connected technology stories.
A survey conducted by Google’s threat intelligence teams of public web content shows that various actors have begun experimenting with indirect prompt injection to target artificial intelligence systems, with the malicious category increasing by 32% between November 2025 and February 2026. Despite the limited sophistication of the observed attacks, Google expects their volume and complexity to grow as artificial intelligence systems become more capable and the cost of carrying out attacks declines.
Google explains its ongoing methodology for combating indirect prompt injection attacks targeting AI applications that use multiple data sources, such as Workspace with Gemini. The methodology relies on detecting and classifying attacks, generating synthetic data, then updating deterministic defenses, machine-learning models, LLMs, and the Gemini model itself.