How WitnessAI raised $58M to solve enterprise AI’s biggest risk | Equity Podcast

Four distinct threat waves emerged since ChatGPT

Enterprise security concerns evolved from employees leaking sensitive data to external chatbots, to blocking harmful AI outputs, to preventing jailbreaks of proprietary models, and now to controlling autonomous agents with inherited human permissions.

Agent authorization creates new attack surfaces

AI agents assume the capabilities and authorizations of their managing users, requiring guardrails at multiple intercept points: the initial prompt, the LLM-generated worklist, and the specific tools the agent attempts to access.

Replacing fear with nuanced business enablement

Unlike traditional cybersecurity that relies on blocking and fear, Witness AI uses natural language policy creation to distinguish context—allowing a garden retailer’s chatbot to discuss 'poison' (weed killer) while blocking the same query at a Manhattan bank.

Behavioral intent detection over blanket bans

The platform can differentiate between an employee searching for internal promotions versus external job opportunities, applying data-loss prevention only when resignation risks are detected rather than blocking all career-related queries.

Network-level interception avoids model vendor conflict

By operating at the network layer rather than within specific LLMs, Witness AI avoids direct competition with OpenAI or Google while integrating with existing security stacks via APIs and SDKs.

Private single-tenant data isolation

Each customer instance runs in their chosen cloud provider, encrypted with their own keys and firewalled from other tenants, ensuring that aggregated AI interaction logs don't create a centralized attack target.

$58M Series A led by Sound Ventures with strategic edge computing backers

The round included Samsung Ventures and Qualcomm Ventures, positioning the company for AI security at the device edge, alongside Fin Capital to expand financial services presence.

Aggressive scaling to match threat velocity

The company expanded from 23 employees at the start of 2025 to 92 currently, opening offices in Mountain View, Atlanta, Asia, and Europe to address what they describe as 'speed running' through traditional enterprise tech evolution cycles.