Onyx Security is an AI security control plane that discovers, monitors, and governs enterprise AI agents across SaaS, cloud, endpoints, and code repositories. Unlike point solutions that address a single layer of AI risk, Onyx provides a unified governance platform spanning observability, security, compliance, orchestration, and ROI measurement.
The company launched in March 2026 with $40M in funding from Conviction and Cyberstarts, emerging from stealth after a year and a half of development. Co-founders Maxim Bar Kogan and Gil Elbaz built the team to 70 people before going public, and the platform was already in use at Fortune 500 companies at launch.
Bar Kogan is a cybersecurity leader and Unit 8200 veteran who won first prize in the 2008 Israeli CodeGuru competition. Elbaz is an AI researcher who previously reported to NVIDIA’s CTO and served in one of the IDF’s AI research units.
What is Onyx Security?
Onyx sits between an organization and its AI agents as a supervisory layer. It discovers both approved and shadow AI across the enterprise, monitors agent actions in real time, and enforces security and compliance policies before agents can execute risky operations.
The platform is powered by its own set of supervisory agents and proprietary AI models built to understand AI reasoning patterns. When a risk is detected, Onyx can block the action, require human approval, or steer the agent in a safer direction.
Key Features
| Feature | Details |
|---|---|
| Agent Discovery | Finds approved and shadow AI across SaaS, cloud, endpoints, and code repositories |
| Guardian Agent | Supervisory AI that identifies and remediates risks automatically |
| Policy Controls | Natural language policy configuration for security and compliance |
| Threat Detection | Real-time analysis of prompts, responses, and agent actions |
| Compliance | EU AI Act, internal governance standards |
| Orchestration | Simplifies agent setup and MCP deployment; optimizes cost, accuracy, and latency |
| ROI Tracking | Adoption metrics, departmental goals, and attainment measurement |
| Scale | 137,000+ agents secured, 593,000+ employees covered, 10M+ sessions analyzed |
| SaaS Coverage | Salesforce, Glean, Microsoft Copilot |
| Cloud Coverage | AWS, Google Cloud, Azure, Oracle |
| Endpoint & Code | Cursor, Claude, GitHub Copilot, GitLab, Bitbucket |
| Human-in-the-Loop | Approval workflows for high-risk agent actions |
How the control plane works
Onyx’s control plane approach means all AI agent activity flows through a single governance layer. The platform discovers agents across the organization’s infrastructure — SaaS applications, cloud environments, developer endpoints, and code repositories — and maps their permissions, data access, and behavioral patterns.
The Guardian Agent continuously monitors these deployments. When it detects an anomaly — excessive permissions, sensitive data exposure, or unauthorized actions — it intervenes based on pre-configured policies. Security teams define policies in natural language rather than code, making the system accessible to non-technical stakeholders.
According to Onyx’s research, 80% of enterprises expose sensitive data through agents, 93% run agents with excessive permissions, and 70% face remote code execution attack exposure. The platform targets these gaps at scale.
AI orchestration and ROI
Beyond security, Onyx handles agent orchestration and business impact measurement. The orchestration layer reduces friction in agent setup and MCP deployment while optimizing for cost, accuracy, and latency. The ROI module tracks adoption metrics and departmental attainment, giving leadership visibility into how AI investments are performing.
Getting Started
When to use Onyx Security
Ideal for enterprises that have adopted or are adopting AI agents at scale and need centralized visibility and governance. Onyx fits best where shadow AI proliferation is a concern — organizations where employees deploy AI tools without security team oversight.
The natural language policy engine makes it accessible to security teams that want governance controls without building custom integrations. The orchestration and ROI layers add value beyond pure security, helping organizations manage the operational and business dimensions of AI adoption.
For more AI security tools and guidance, see the AI security tools category page. For runtime prompt protection, see Lakera Guard or LLM Guard. For LLM vulnerability scanning, look at Garak or Promptfoo. For AI agent access control, see Alter. For zero trust enforcement at the protocol layer, check Xage Security.
