October 20, 2026: WitnessAI Release

ChatGPT Dots is now supported for prompt and response observability. Activity appears in the console as ChatGPT Dots, ChatGPT Dots Enterprise, or ChatGPT Dots Personal, depending on the account type.

Fireworks AI is now supported, with policy enforcement on prompts and observability for tool calls.

The Ask Claude assistant on the Claude Platform documentation site (platform.claude.com/docs) is now supported for prompt and tool call observability.

The policy editor's Sources & Destinations tab is now called Rules and presents each policy as a single readable rule: IF these sources, AND these destinations, THEN Allow or Block. Sources, Destinations, and Controls are collapsible sections, selected values appear as removable tags, and you can search for values, including searching all users by email. The Block/Guardrail choice has moved from the General tab into Controls. Existing policies open and save with their current settings unchanged.

Discovery now has its own second-level navigation with three pages: Applications, Agents, and MCP Servers. Applications contains the Discovered Apps and App Catalog tabs, and MCP Servers contains the Discovered MCP and MCP Catalog tabs, replacing the previous view toggle.

In the Alerts policy column, guardrails that did not fire are now visually dimmed, and hovering a guardrail icon shows whether it fired. Guardrail icons have also been refreshed for consistent appearance across the console.

No Performance, Reliability & API Updates in this release.

Fixed an issue where prompts from Claude Code in the browser could be categorized as Claude Cowork Enterprise in the console.

When you select See All on the Home dashboard Discovery widget with Last 30 Days selected, Discovery may open with the default Last 7 Days filter instead.

Application names may in some cases overwrite one another on a user's tab.

On Warning alerts, the Allowed action badge may not be displayed alongside the Warning action.

For guardrails with large test sets, the Refinement step may fail with an error, and Validation may count some test cases as failed without evaluating them, lowering the displayed accuracy.

The slider in the AI test generation view may not move proportionally.

No Breaking Changes in this release.

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