AI “Stealth” Technology Moves Beyond the Digital World

Researchers at Cranfield University have developed a novel technology that enables physical objects and communication signals to evade detection by artificial intelligence systems, marking a significant step forward in red teaming future AI-enabled critical systems.

The innovation, now the subject of a patent pending publication with amendments at the Great Britain IP Office and the World Intellectual Property Organisation (WIPO – WO2026041737), introduces a new approach to manipulating physical entities in ways that remain undetectable to a wide range of AI models.

At its core, the technology uses generative AI to iteratively modify objects or signals until they can bypass AI-based detection systems, while still preserving key aspects of physical realism. This moves beyond traditional approaches focused on digital image or video manipulation, extending AI “stealth” capabilities into real-world environments.

The work has been developed over several years through two major EPSRC-funded initiatives: the Trustworthy Autonomous Systems – Security Node and the 6G Future Communications Hub, CHEDDAR. By bridging advances in AI, security, and next-generation communications, the research highlights emerging challenges and opportunities at the intersection of intelligent systems and physical infrastructure.

The technology has already attracted interest from industry, reflecting its potential applications across sectors where AI-driven sensing and detection are becoming increasingly central.

The patent has been led by researchers Weisi Guo, Chen Li, and Antonios Tsourdos at Cranfield University.