The emergence of autonomous agents presents cybersecurity teams with an unprecedented challenge, with systems capable of interacting with each other, calling internal or external models and triggering operational actions. The rapid growth of these uses requires us to understand which AI entities are really active, how they communicate and what permissions are granted to them.
Sweet Security is developing a platform that maps all of the AI agents, models and services deployed within a company’s systems. This approach aims to provide comprehensive visibility of flows and behaviors, in a context where an increasing share of risks comes from AI services deployed without clear governance or connected in an uncontrolled manner to sensitive environments.
The platform identifies the models used, internal LLM servers, active agents and application integrations. It analyzes the permissions assigned to agents and detects configurations that could create vulnerabilities. It also reveals the presence of Shadow AI, undeclared solutions, often created or connected by business teams, which escape traditional security controls.
Sweet Security highlights an approach focused on the visibility of executions and behaviors, whether it concerns application flows, access to critical resources or cross-model integrations. The goal is to enable security leaders to focus on actions that have a tangible impact on production environments.
Sweet Security raised $75 million, or €65 million, in Series B, led by Evolution Equity Partners with the participation of Munich Re Ventures, Glilot Capital Partners and Key1 Capital. The company, founded by Dror Kashti, former CISO of the Israeli Armed Forces, Eyal Fisher, former head of Unit 8200’s cyber department, and Orel Ben-Ishay, former head of R&D in Special Operations, brings its total funding to $120 million. It employs around 70 people in Israel and the United States.
Several players are emerging alongside Sweet Security. Zenity offers an AI agent governance and observability platform spanning SaaS, cloud and desktop. Likewise, Cycode recently launched an “AI & ML Inventory” module aimed at eliminating the blind spot of Shadow AI. Finally, Obsidian Security is interested in the discovery and management of unauthorized uses of generative AI applications.
