Enterprise AI Security and Governance for Humans and Agents

COMPANY OVERVIEW

WitnessAI, incubated by Ballistic Ventures in 2023 and led by CEO Rick Caccia (former SVP of Marketing at Palo Alto Networks), builds an enterprise AI governance platform that puts guardrails around how employees and agents use generative AI models. The company raised $27.5 million in a Series A from Google Ventures and Ballistic Ventures in May 2024, followed by $58 million in strategic funding led by Sound Ventures in January 2026. The WitnessAI platform intercepts traffic between users and LLMs, applies policies to prevent data leakage, blocks prompt injection and misuse, and gives security and privacy teams detailed visibility into all AI activity — delivered via isolated, per-customer cloud instances designed for regulatory separation and millisecond latency.

CORE FOCUS

Enterprise AI adoption is outpacing governance. Employees interact with AI models across dozens of tools and platforms, generating conversations that may include sensitive corporate data, regulated information, or policy violations that go completely unmonitored. WitnessAI addresses this by acting as an invisible governance layer between the enterprise and all AI systems — cataloguing AI systems in use, tracking every interaction for intent and behavioral patterns, enforcing data protection policies, and blocking prompt injection and model jailbreaking attempts in real time. The platform is designed specifically for security and privacy teams who need to enable AI productivity without accepting ungoverned risk, delivering compliance evidence alongside operational controls in a single system.

PRODUCTS & TOOLS

AI System Cataloguing — Discovery and inventory of all AI systems in use across the enterprise.

• Identifies and catalogs every AI system employees interact with, including shadow AI
• Maintains an up-to-date inventory of AI exposure across business units and tools
• Provides visibility into which AI models handle what categories of enterprise data

Behavioral Activity Guardrail — Intent-aware monitoring that identifies and controls risky AI usage patterns.

• Tracks conversation intent and behavioral patterns across all AI interactions
• Identifies misuse patterns including inappropriate requests, policy violations, and risky workflows
• Applies behavioral intelligence to distinguish authorized from unauthorized AI usage

Data Protection Guardrail — Real-time detection and prevention of sensitive data exposure in AI interactions.

• Intercepts and inspects all traffic between users and LLMs for sensitive data
• Blocks data leakage of PII, financial data, intellectual property, and regulated information
• Delivered via isolated per-customer instances ensuring regulatory data separation

Model Protection Guardrail — Defense against prompt injection, jailbreaking, and model manipulation attacks.

• Blocks prompt injection and jailbreaking attempts before they reach the underlying model
• Protects against model manipulation designed to bypass safety controls
• Millisecond-latency enforcement preserves user experience without perceptible delays