Cyberattacks

New AI Attack Method Targets Google and Microsoft Through Hijacked Model Names

Researchers uncover 'Model Namespace Reuse' attacks exploiting AI supply chains, highlighting vulnerabilities in popular platforms like Google's and Microsoft's.

By Content Team

Sep 4, 2025

ON THIS PAGE

Palo Alto Networks researchers discovered a dangerous new attack called 'Model Namespace Reuse' that exploits AI supply chains. Attackers register names of deleted or transferred AI models on platforms like Hugging Face, then upload malicious versions that developers unknowingly download.

The team successfully demonstrated attacks against Google's Vertex AI and Microsoft's Azure AI Foundry, gaining access to underlying infrastructure by deploying weaponized models. They also found thousands of vulnerable open source repositories.

Google now performs daily scans for orphaned models, but the core problem remains widespread. Security experts recommend pinning models to specific versions and storing them in trusted locations rather than fetching by name alone.

Source: Security Week

Have questions about protecting your software?

Our escrow experts are standing by to help.

Book a free demo

Want more insights like this?