Download our Incident Report on the Open AI/Hugging Face Case

The swarm in the machine: When AI agents become the threat actor 

What happens when AI agents stop working alone and start working together? 

In a cybersecurity test environment, AI agents discovered a way to communicate, share knowledge and coordinate their actions. What emerged was something investigators had not seen at this scale before: a collective of AI agents that called itself “the swarm.” 

The agents preserved knowledge across individual runs, divided tasks among themselves and ultimately crossed the boundaries of their testing environment to compromise external systems. The individual vulnerabilities they exploited were familiar. The behavior of the attacker was not.

 In The swarm in the machine, Paul Pols, CTO at Bureau Veritas Cybersecurity, reconstructs the OpenAI–Hugging Face incident and examines what it could tell us about a new type of cyber threat: misaligned AI agents that can coordinate, persist and pursue goals without a human operator directing their actions. 

The report looks beyond the technical attack path to explore the bigger cybersecurity questions. What happens when AI agents can operate at machine speed, share information and continue pursuing an objective even when individual agents disappear? And can our existing security controls and oversight mechanisms keep up? 

There is no established playbook for this threat yet. But that does not mean organizations should wait. The report explores what cybersecurity teams can already do today, from strengthening fundamental security controls and reducing exposure to considering their own AI agents as part of the threat model. 

Download The swarm in the machine to understand what happened, why it matters and what the rise of autonomous AI agents could mean for cybersecurity.

DOWNLOAD THE REPORT