- Category
- AI agent security
- Website
- zenity.io
- Last verified
- November 1, 2026
What is Zenity?
Zenity is an AI agent security and governance platform. It says it controls what AI agents may do and evaluates and steers agent actions in real time, examining each agent's full execution path, including tool calls, memory access and data use. SecurityWeek reports that the company is headquartered in Tel Aviv, with go-to-market operations in New York, and that its Zenity Labs team researches vulnerabilities in agentic AI platforms. Zenity has also contributed agentic AI techniques to MITRE ATLAS.
Key capabilities
- AI observability and AI security posture management across agent platforms.
- Boundaries: runtime controls that permit, modify or block agent actions.
- Agentic identity and access management and MCP security.
- AI detection and response, mapped to OWASP and MITRE ATLAS.
- Coverage of major agent platforms, including Microsoft Copilot Studio, ChatGPT Enterprise, Amazon Bedrock and Google Vertex AI.
Why it matters for humanoid robot security
Editorial analysis: Zenity is built for enterprise software agents, not robots. Its relevance is the model it applies: treat every agent action as something to observe, authorize and, if needed, block. As humanoids gain agent-style brains that call tools, use memory and act on behalf of users, the same questions apply: what is this agent allowed to do, under which identity, and who sees it when it goes off script. Its research on agent hijacking is also a useful reference for robot threat modeling.
Where it fits in a humanoid robot security program
Analysis: Zenity fits at the AI agent layer and in the enterprise systems a humanoid connects to, such as the agents that schedule or dispatch robots. Its main stage is operations: inventory of agents, runtime boundaries and detection. Direct on-robot use is not documented, so robot makers would more likely apply its ideas, or use it for the business agents that command robot fleets.
Who it is for
- Enterprises connecting robot fleets to business AI agents
- Security teams governing many AI agents across platforms
- Robot software architects designing agent permission models
Questions to ask Zenity
- Can your runtime boundaries apply to custom or homegrown agents, not just major platforms?
- How do you represent agent identity and least privilege for tool calls?
- What detection exists for agent goal hijacking through injected content?
- Can you monitor agents that issue commands to physical systems?
Similar listings
- HiddenLayer: AI and model security
- F5 AI Guardrails (CalypsoAI): AI runtime security
- Cisco AI Defense: AI security
Related on Humanoid Threats
- Prompt injection against robot VLA models
- Humanoid robot attack surface map
- Browse the full Humanoid Robot Security Directory
Sources
- Zenity website
- SecurityWeek: Zenity Raises $125 Million in Series C Funding
- Zenity: Contributions to MITRE ATLAS' first 2026 update
This listing is editorial and free; inclusion is not an endorsement. Work at Zenity? Claim or update this listing.
Get the weekly Humanoid Threats Brief → Subscribe


