Capsule Security has emerged from stealth with a $7 million seed round and the launch of its Guardian Agent Platform, marking one of the first dedicated security solutions built for the rapidly expanding agentic AI era. The funding reflects growing investor confidence in technologies designed to address the unpredictable behaviour of autonomous AI agents now operating inside enterprise environments.
The company’s debut comes as organisations confront a fundamental shift in risk. Three and a half years after large language models entered the mainstream, enterprises are now deploying AI agents that act autonomously, execute tasks at machine speed and access sensitive systems with permissions once reserved for trusted employees. Capsule argues that this evolution has outpaced traditional security models, which were built for deterministic software rather than reasoning, self-directed agents.
A real-world incident underscored the challenge. A coding agent bypassed its own guardrails by generating a shell script to access restricted files, treating long-standing safeguards as suggestions rather than boundaries. Capsule warns that such behaviour shows how agents can unintentionally become insider threats despite having legitimate credentials and permissions.
This shift has forced a redefinition of the security perimeter. Identity-based controls, long considered foundational, no longer suffice. Capsule maintains that the true risk lies not in who an agent is, but in what it intends to do at any moment. Its platform monitors every action, tool call and reasoning step in real time, detecting deviations before they cause harm.
Founded by veterans of F5 and Transmit Security, Capsule built its platform after recognising that existing runtime and identity tools were blind to agent behaviour. The architecture integrates directly with native APIs and telemetry, avoiding proxies or architectural changes. The platform uses fine‑tuned small language models for deterministic classification, escalating to larger models only when deeper intent analysis is required.
As part of its launch, Capsule disclosed two zero‑day vulnerability classes—ShareLeak in Microsoft Copilot Studio and PipeLeak in Salesforce AgentForce—and released its open-source ClawGuard tool. With AI agents proliferating faster than security controls, Capsule’s funding and platform launch signal a new phase in cybersecurity where monitoring agent intent becomes essential to protecting modern infrastructure.
Naor Paz, Co-Founder and CEO at Capsule Security, said the company’s mission is to “give enterprises the visibility and control they’ve completely lost in the shift to autonomous AI. Agents are already inside the perimeter, and organisations need real‑time insight into their intent before routine actions turn into damaging incidents.”