March 19, 2026 WitnessAI Update

💡 Note: This release adds support for OpenAI’s newer Responses API, introduces custom data types for Data Protection, improves how WitnessAI classifies Claude usage, and resolves several customer-reported issues across integrations, policy enforcement, and console reliability.

New Features

Custom Data Types for Data Protection

Policy authors can now define custom sensitive data types within the Data Protection guardrail. This extends detection coverage beyond built-in data types, allowing organizations to tailor data protection policies to their specific compliance and security requirements. Custom data types are configurable in the Policy settings.

OpenAI Responses API Support (Limited Availability)

WitnessAI now supports capturing traffic from the OpenAI Responses API (/v1/responses). Conversations and agentic activity using this newer OpenAI endpoint will now appear in the console alongside other AI activity. This capability is currently available on request.

Feature Updates

Claude: Personal vs. Enterprise Usage Differentiation

WitnessAI now distinguishes between personal and enterprise Claude usage. The console will correctly classify and surface Claude activity based on whether the session is personal or enterprise-authenticated, enabling more precise policy enforcement and reporting.

Conversations: Pagination Added to Session Content Panel

The session conversation content panel now includes pagination controls, making it easier to navigate long conversations. Pagination improves usability for high-volume sessions.

Data Protection: Australian Identity Data Types Added

Detection coverage for Australia-specific identity data types has been added to the Data Protection guardrail. Organizations monitoring activity in Australia can now apply policies covering local identity formats such as tax file numbers and Medicare card numbers.

App Catalog: Sync Functionality Updated

App Catalog sync functionality has been updated and consolidated. Sync behavior between the catalog and downstream services is now more reliable and consistent.

Known Issues

No known issues identified in this release.

Fixed Issues

SECURITY

Policy enforcement for private applications is now fully reliable. In some cases, policies targeting Private Application destinations were not saving as expected. This has been resolved.

Identity management: user counts now stay in sync. In some cases, user counts in the console were not reflecting the correct total from the connected identity provider. This has been resolved.

INTEGRATIONS

Microsoft 365: Dictation works seamlessly alongside WitnessAI. In some configurations, the Microsoft 365 Dictation feature in Word and Copilot was not functioning as expected while WitnessAI was active. This has been resolved.

Prompt capture is now more reliable across AI applications. In certain configurations, prompts were not appearing in the console as expected. This has been resolved and activity is now consistently captured and visible.

Agentic tool calls now correctly show when an error has occurred. In some cases, Tool Call cards were not displaying an error state when expected. Cards now correctly reflect error conditions.

PLATFORM RELIABILITY

Dashboard: “Users of AI Applications” widget now shows accurate counts. In some environments, this widget was not displaying counts accurately for larger user populations. This has been resolved.

Alerts & Reporting: Latency data is now more accurate. Latency reporting data was not always reflecting expected accuracy in some conditions. This has been resolved, improving the reliability of latency data in the console.

Agentic activity views are cleaner and more accurate. In some cases, connector actions in conversation details and the Discovery → Agents view were appearing more than once for the same event. Entries are now displayed correctly.

Console responsiveness improved during periods of high activity. Processing optimizations have been applied to reduce delays and keep the console responsive under load.

Breaking Changes

No breaking changes identified in this release.

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