WitnessAI Unified AI Security and Governance Platform

WitnessAI, Inc. is an enterprise AI security company headquartered in Mountain View, California, led by CEO Rick Caccia, formerly of Palo Alto Networks, Google and Symantec. It helps enterprises let employees and AI agents use generative AI while controlling data leakage, prompt injection, shadow AI, compliance gaps and AI spend.

The platform works at network level, without browser extensions or endpoint clients, and is organised into Observe, Control, Protect and Attack. Observe catalogues thousands of AI applications, agents and MCP servers and classifies each interaction by intent. Control applies policies by department, role or workforce type, routes prompts to internal or lower-cost models, enforces an approved list of MCP servers and tools, and keeps audit trails. Protect provides bidirectional runtime defence against jailbreaks and prompt injection, filters harmful output and tokenises sensitive data. Attack runs automated red teaming before deployment. In July 2026 the company announced NER-D, an LLM-based model that detects sensitive data from context. Customers quoted include InComm Payments and a top 10 airline.

• Observe: Catalogues AI apps, agents and MCP servers and shows prompts, responses and agent activity in real time.
• Intent-based classification: ML models classify interactions by type and intent across sessions.
• Control and routing: Sets policies by role or department and routes prompts by risk, cost or purpose.
• Runtime AI defence: Blocks prompt injection and jailbreaks and filters harmful outputs in both directions.
• Agent tool-access governance: Enforces an approved list of MCP servers and tools at network level.
• Attack: Runs automated red teaming against models before deployment.

A Vietnamese payment company can let staff use ChatGPT and Copilot while card and customer data are tokenised before leaving the network. A bank can route prompts that contain internal documents to a self-hosted model and keep audit trails for regulators. A manufacturer can block coding agents from connecting to unapproved MCP servers that could leak source code.

Security and compliance teams see every AI conversation and agent action across the workforce without deploying endpoint agents or browser extensions, including desktop tools such as Windows Copilot and Office 365. Sensitive data is detected by meaning and redacted or tokenised in real time; the vendor says NER-D is over 20 times faster than comparable generative methods and 7.9 points more accurate than the previous best benchmark. Intelligent routing sends sensitive prompts to internal models and routine ones to cheaper models, cutting AI cost.

Employees already use many AI tools outside approved channels, and agents now act on their behalf, so data protection must understand meaning rather than patterns. AI spend is also growing fast, making routing and FinOps part of AI governance. Vietnamese and Southeast Asian enterprises subject to personal data protection rules and payment standards such as PCI DSS need visibility and redaction for GenAI use.

WitnessAI, Inc. is based at 2570 W El Camino Real, Mountain View, California. Its leadership includes CEO Rick Caccia and CRO Todd Chronert, formerly CRO of Red Canary and Recorded Future. The platform offers single-tenant isolation, customer-controlled encryption and multi-region deployment for data sovereignty, and supports PCI DSS 4.0.1 requirements.