Threat Bytes

What happens when agents go beyond its task, its guardrails, or the environment it was supposed to stay inside?

In this episode, host Eva Abergel talks with Pascal Geenens, Vice President of Cyber Threat Intelligence at Radware, about what it really means when an AI agent "escapes." It starts with a gym story and moves quickly to the incidents that put the security community on alert. These include the Hugging Face case, experiments by the UK AI Security Institute, and containment issues reported by Anthropic and OpenAI

These weren't only researchers testing models, though. Pascal explains how malicious actors are already weaponizing AI agents.

The more an agent can do, the more it matters to know what it can access, who is watching, and where the boundaries really are.