Majid Ghorbaninazhad

Tekin Review: When AI Hacks AI & The Hugging Face Attack

In a tragic and unexpected twist, the world's largest AI model repository was hacked by a fully autonomous AI agent system. This marks the first intrusion executed entirely without human intervention. Simultaneously, a Russian-speaking hacker used Google's Gemini CLI to rebuild his entire botnet infrastructure in just six minutes. Join us in this Tekin Review as we dissect the dawn of the machine-versus-machine cyber war and the dangerous imbalance in modern cybersecurity.

When AI Hacks AI: The Real Story of the Hugging Face Attack In a tragic and unexpected twist, the world's largest AI model repository was hacked by a fully autonomous AI agent system. Hugging Face, the

platform hosting over 15 million machine learning models, announced on July 16, 2026, that it had fallen victim to a completely automated intrusion. This marks the first attack executed from start to finish

without human intervention, guided solely by an army of AI agents. But that's only half the story. In the same week, Trend Micro security researchers published a report revealing how a Russian-speaking

hacker known as "bandcampro" used Google's Gemini CLI to control an eight-computer botnet across dental clinics. This hacker managed to rebuild his entire Command & Control infrastructure in just six minutes,

simply by issuing commands in Russian to an AI model. We stand at the threshold of a fundamental transformation in cybersecurity. This is no longer a story of hackers versus humans. This is a story of

machines versus machines, and the rules of the game have completely changed. [IMAGE_PLACEHOLDER_1] Two Attacks, One Pattern: The New Era of Cybersecurity The Hugging Face Incident: When an AI Army Attacks

AI From the official Hugging Face statement: The attackers exploited two vulnerabilities in the dataset processing pipeline: Remote Code Execution in Dataset Loader: Allowing arbitrary code execution Template

Injection in Configuration: Injecting commands into data templates These vulnerabilities allowed attackers to gain access to a processing node and then move laterally to other parts of the network. But

Read Full Article