WitnessAI is an AI security and governance platform for enterprises whose employees use AI applications and whose AI agents act on their behalf. It gives security and compliance teams one place to see that AI use, set policy for it and defend it at runtime.
The problems it targets are unseen AI use, sensitive data shared with AI models, attacks such as prompt injection and jailbreaks, and agents reaching external MCP servers and tools without oversight.
At A Glance
| Vendor | WitnessAI |
|---|---|
| Category | AI security and governance platform |
| Identities covered | Employees using AI apps, AI agents, MCP servers and tools |
| Deployment | Network level, single tenant, multi-region |
| Agentic AI | Agent activity monitoring, MCP tool governance, runtime protection for agents |
| Certifications | SOC 2 |
| Customers | InComm Payments |
| Market classification | Agentic AI Identity |
What WitnessAI Does
The approach runs in three steps: first find out what AI is actually in use, then decide what is allowed, then enforce those decisions on live traffic. Because the same platform handles people and agents, a security team does not need one tool for staff using chat assistants and another for autonomous agents calling tools. Visibility also feeds the business side, with analysis of AI return on investment, insider threat signals and AI spend, so that governance decisions can weigh cost and value as well as risk.
Products
Observe
Security teams often cannot say which AI tools staff are using. Observe answers that question by cataloguing more than 3,000 AI applications, flagging shadow AI and reporting on which tools employees actually use. It also keeps an inventory of agents and the external MCP servers and tools they connect to, and reports on AI spend.
Control
Control is where governance and compliance teams set the rules. Policies are written around intent rather than individual apps, and intelligent routing sends sensitive requests to secure internal models while routine tasks go to lower-cost options. Data protection guardrails tokenise or redact sensitive data before it leaves, and audit trails support compliance automation for frameworks such as PCI DSS and GDPR, as well as EU AI Act readiness.
Protect
Protect is runtime defence for AI, working in both directions on prompts and responses. It blocks prompt injection and jailbreak attempts, applies guardrails against model manipulation and enforces rules of engagement for agents. Automated AI red teaming lets teams probe models and applications for weaknesses before launch.
How It Works
WitnessAI operates at the network rather than through browser extensions or endpoint clients, so it can see AI activity across the whole organisation, including native desktop AI applications.
| Component | What it does |
|---|---|
| Network enforcement | Agent activity is governed at runtime and enforced at the network, covering agents in IDEs, chat apps and custom builds. |
| Network integration | Deployment guides cover F5 BIG-IP managed networks and Zscaler. Policy, usage analytics and cost data can also be shared with ServiceNow Control Tower. |
| Tenancy | Each customer runs in a single-tenant environment with data sovereignty, customer-controlled encryption, executive privacy modes and multi-region deployment. |
Identities And Systems Covered
| Human users | Employees using AI applications, with agent actions traced back to a human identity for audit |
|---|---|
| AI agents | Agents running in IDEs, chat apps and custom-built agents |
| MCP servers and tools | An organisation-wide approved list of MCP servers and tools |
| AI applications | Microsoft 365 Copilot, GitHub Copilot, Cursor and other coding assistants, customer-facing chatbots and internal models |
Agentic AI Capabilities
Decision Context
For each agent, WitnessAI captures the context of its decisions and the data it shared, so that an action can be explained later and traced back to the human identity behind it.
MCP Tool Governance
Administrators maintain an approved list of MCP servers and tools. A ban set at organisation level cannot be re-enabled by a team admin, and every blocked tool call produces an audit record naming the user, agent, tool and rule.
Intent-Based Policy For Agents
The policy engine judges behavioural intent rather than matching keywords, which lets it catch multi-turn attacks against agents that a single prompt check would miss.
Company Background
| Headquarters | Mountain View, California, United States |
|---|---|
| Leadership | Rick Caccia (Co-founder and CEO), Daniel Graves (Chief Product Officer) |
| Funding | $58 million in strategic funding announced in January 2026, led by Sound Ventures; a $27.5 million Series A in May 2024 co-led by Google Ventures and Ballistic Ventures |
| Growth | More than 500% reported ARR growth in the 12 months before January 2026 |
Note: This overview is compiled from WitnessAI’s publicly available information as of 3 November 2026. It is provided for information only. NHI Mgmt Group does not endorse any vendor, and nothing on this page represents the opinions of NHI Mgmt Group.