WitnessAI

There's a scene playing out in IT security departments everywhere right now. A CISO opens a dashboard and sees hundreds of employees using AI tools the company never approved. Somewhere in those conversations, financial data was pasted into a chatbot. A contractor asked an AI agent to search internal documents. A developer's coding assistant quietly connected to a third-party MCP server. No one noticed. Nothing stopped it.

WitnessAI noticed. And it stopped it.

Founded in 2023 and incubated within Ballistic Ventures - a VC firm that exclusively bets on cybersecurity - WitnessAI built a platform around a single uncomfortable truth: enterprises are deploying AI faster than they can govern it. The product operates at the network layer, cataloging every AI interaction across employees, models, applications, and agents, then enforcing behavioral policies in real time. No code changes. No browser extensions. No gaps.

Remember shadow IT? Employees spinning up cloud tools the company never approved, creating data sprawl and compliance nightmares? That problem took a decade to manage. Shadow AI is moving on a faster timeline, and the consequences are different in kind, not just degree.

When an employee pastes customer PII into ChatGPT, they're not circumventing a file-sharing policy. They may be violating GDPR, PCI DSS, or industry-specific regulations - without knowing it, and without leaving a trace in any existing security tool. Traditional firewalls see encrypted traffic as a tunnel. DLP tools focus on files. SIEM platforms log events after the fact. None of them were built to understand what a prompt actually means, or what an AI agent is actually doing.

WitnessAI decided the answer wasn't to bolt AI governance onto existing security tools. It was to build a new category from scratch - one that understands intent, not just patterns.

Rick Caccia spent years at Palo Alto Networks building security products before spending time at Google and Red Canary. Gil Spencer had already built and sold two security companies - IronKey (hardware-encrypted USB drives, acquired by Imation) and Marble Security (mobile security, acquired by Proofpoint) - plus a stint engineering Mac software at Apple in the Quicktime era. When they co-founded WitnessAI, they weren't guessing at what enterprises needed. They'd watched the playbook fail before.

The bet: enterprises won't slow down AI adoption, and telling them "no" isn't a product. The only winning strategy is a platform that makes "yes" safe.

The financial services sector, utilities, airlines, automotive, retail, and telecoms share something besides good lawyers: they cannot afford an AI incident. Customer data, regulated transactions, proprietary models - the downside of getting AI governance wrong is not an awkward press release. It is regulatory fines, customer loss, and potential liability.

SK Telecom, one of South Korea's largest telecom carriers, deployed WitnessAI to secure GPT-4 implementations handling Korean-language customer conversations. Telecommunications is a category where customer data flows at enormous scale and where one prompt injection could expose millions of records. That's not a beta customer. That's a proof point.

The company's 500% ARR growth in 2025 and 5x headcount expansion suggest the proof point is not an anomaly. Strategic investors Qualcomm Ventures and Samsung Ventures - both with direct interests in enterprise AI deployment at scale - joined the Series B in January 2026, adding industrial weight to the financial backing of Sound Ventures and existing investors Google Ventures and Ballistic Ventures.

AI chatbots were the opening act. The real transformation is AI agents - systems that don't just answer questions but take actions: querying databases, writing code, sending emails, booking meetings, executing API calls. Enterprises are already deploying them. The Model Context Protocol (MCP) is connecting agents to tools at a rate that makes 2024's shadow AI problem look quaint.

WitnessAI launched agentic security capabilities in January 2026 specifically because the governance gap isn't shrinking - it's widening. The platform now fingerprints MCP servers, categorizes their functional capabilities (software development, research, data access), tracks tool access in real time, and creates a human-agent identity connection so every autonomous action has a traceable owner. When an AI agent takes action on behalf of an employee, WitnessAI treats it as an extension of that employee's access policy.

That design choice matters. Most security tools govern people. WitnessAI governs both people and the AI acting on their behalf from the same policy engine. As AI agents become a core part of the enterprise workforce, that distinction becomes the entire product.

Here's what the opening scene looks like now: a CISO opens the WitnessAI console. The dashboard shows every AI interaction across the organization, categorized by risk, flagged by policy, traceable to an identity. An agent tried to connect to an unapproved MCP server at 2:47 AM. The connection was blocked automatically. The incident report is already generated. Nobody had to notice because the platform noticed for them.

That's not a vision statement. It's a product demo. The gap between enterprise AI ambition and enterprise AI governance is exactly the gap WitnessAI was built to close - and it's closing it one blocked prompt at a time.