June 16, 2026 WitnessAI Release

The AI Value Dashboard gives administrators a financial view of the value WitnessAI delivers across the organization. The dashboard surfaces total risk avoided and cost savings for the period, broken down by categories: prompts blocked, data leaks prevented, and proprietary information leaks prevented. The AI Impact score shows the percentage of productive AI usage. The Risk Distribution chart and Risk Events trend chart provides at-a-glance visibility into where risk is concentrated. Administrators can configure custom keyword sets to classify usage as productive or non-productive, and a new Proprietary Information Escalation Rate metric is also included. Settings are managed in the ROI config drawer. Phase 2, including drill downs and department filters, is planned for a future release.

Agentic Action Controls enable administrators to view and control which MCP Servers and individual MCP Tools are available vs. blocked on the Discovery → MCP Servers, Discovery → Agents, and Policy consoles.

The Employee AI Agents table on the Supported Applications & LLMs documentation page shows which Agents are currently supported for Agentic Action Control

WitnessAI now supports Claude Code Browser (claude.ai/code), Anthropic's browser-based agentic coding environment. Conversations are captured and visible in the Conversations console under the Software Development category. In-app warnings, blocking, and redaction are not yet available for Claude Code Browser.

Claude Teams accounts (claude.ai team plans) are now recognized as a distinct application type, separate from Claude Personal and Claude Enterprise. Administrators can apply independent routing rules and policy targeting to Teams-tier accounts.

No feature updates in this release.

No Performance, Reliability, & API Updates in this release.

Fixed an issue where personal-tier Claude.ai traffic could in some cases be missed following Anthropic's migration to versioned API endpoints, which could result in prompts not being recorded and policies not being enforced.

Fixed an issue that could cause the context service to become unstable on long multi-turn ChatGPT Enterprise conversations.

Fixed an issue where app usage event recording could occasionally deadlock under high concurrency, resulting in gaps in usage data.

Fixed an issue where the conversation details flyout could show incorrect prompt counts or display "Unknown User" for correctly attributed sessions.

Fixed an issue where agentic conversation sessions could display "Unknown User" in the console even when the user was correctly linked.

Fixed an issue where users with the console-user role could occasionally see a "Failed to fetch policies" error banner on the Alerts page.

Fixed an issue where prompts containing certain special characters could in some cases be dropped and not recorded.

No known issues in this release.

No breaking changes in this release.

Please contact Customer Success if you have questions or need support with this release.