Capsule Security
Category: AI in Cybersecurity
Runtime security trust layer for AI agents, preventing manipulation, data exfiltration, and unsafe behavior in real time Capsule Security was founded in 2025. The company is led by Naor Paz. Based in Tel Aviv, Israel. Team size: 11-50. Total funding raised: $7M. Latest round: Seed. Key investors include Lama Partners, Forgepoint Capital International.
- Founded
- 2025
- Headquarters
- Tel Aviv, Israel
- Team size
- 11-50
- Total funding
- $7M
Value proposition
Real-time runtime security layer that monitors, detects, and blocks unsafe AI agent behavior before actions execute — without requiring code changes, proxies, gateways, or SDKs
Products and solutions
Capsule Guardian Agent (multi-agent runtime security system), ClawGuard (open-source pre-invocation checkpoint tool), Agent Security Graph, Runtime Protection, Deep Observability, Agent Identity Control, Whitebox Red Teaming
Unique value
Only platform providing runtime intervention (not just detection) for AI agent behavior, using fine-tuned SLMs as "Guardian Agents" to protect AI with AI, with frictionless deployment across any agent framework
Target customer
Enterprise security teams deploying AI agents (coding agents, SaaS agents, enterprise agent platforms)
Industries served
Cybersecurity, Enterprise IT, Financial Services, Technology
Technology advantage
Fine-tuned Small Language Models (SLMs) powering a multi-agent "Guardian Agent" system; Agent Security Graph mapping agent-tool-data relationships; real-time runtime intervention capability; discovered and disclosed zero-day vulnerabilities in Microsoft Copilot Studio (CVE-2026-21520) and Salesforce Agentforce
How they differentiate
Unlike traditional security tools that only audit after the fact, Capsule enforces runtime controls inside the execution path, blocking unsafe commands before they execute. Uses fine-tuned Small Language Models (SLMs) as a multi-agent guardian system. No proxies, gateways, SDKs, or browser extensions required. Works across all major agent platforms (Cursor, Claude Code, Copilot Studio, ServiceNow, Salesforce Agentforce).
Main competitors
CalypsoAI, Trent AI, Protect AI, Lasso Security
Key partnerships
CrowdStrike Startup Accelerator (finalist, alongside AWS and NVIDIA), Advisory board includes Chris Krebs (former CISA Director), Omer Grossman (ex-CyberArk Global CIO), Jim Routh (former Fortune 500 CISO), Dr. Yonesy Núñez, Agent Control Standard (ACS) — co-founded open standard for AI agent runtime governance, Gartner — named Representative Vendor in inaugural Market Guide for Guardian Agents
Major milestones
April 2026: Emerged from stealth with $7M seed round, Disclosed ShareLeak (CVE-2026-21520) vulnerability in Microsoft Copilot Studio and PipeLeak vulnerability in Salesforce Agentforce, Released open-source ClawGuard tool, Named finalist in CrowdStrike/AWS/NVIDIA Cybersecurity Startup Accelerator (out of ~1,000 applicants), May 2026: Launched Agent Control Standard (ACS) — open standard for runtime governance of AI agents — and hooks.security catalog, June 2026: Named Representative Vendor in Gartner inaugural Market Guide for Guardian Agents
Market positioning
Early-stage leader in the emerging AI agent runtime security category, positioned at the intersection of cybersecurity and agentic AI. Differentiated by runtime intervention (vs. post-hoc detection) and frictionless deployment.
Geographic focus
Global (headquartered in Israel, targeting enterprise customers worldwide)
About Naor Paz
Ex-F5; 8 years in Israeli intelligence community including Unit 8200 and Security Research & Analytics Team Lead at Israeli Prime Minister's Office; 15+ years in tech/security
Official website: https://www.capsulesecurity.io/