Agents release notes

PII in-flight masking now extends to files that are dynamically downloaded by agents during runtime and passed as multi-modal inputs to LLMs. Previously, PII masking only covered text input routed through LLMs during agent execution. Agents that retrieve and process documents — such as PDFs, spreadsheets, or images — now receive the same level of data protection.

When an agent downloads a file and passes it to an LLM, the system extracts the file content, detects and pseudonymizes PII entities, and sends only the masked content to the model. For image files, Optical Character Recognition (OCR) is applied to extract text before detection. After the LLM returns a response, placeholders are rehydrated with the original values in the output.

No additional configuration is required. If PII masking is enabled for agents in your AI Trust Layer policy, it automatically applies to both text inputs and dynamically downloaded files.

For details, refer to Analyze files and PII masking.

Correction, June 8, 2026: This feature was released on June 5, 2026, but was not included in the release notes at the time.

The Prompt editor is a new rich text editing experience for writing agent prompts. It replaces the previous plain-text editor with a structured interface where input arguments, tools, contexts, escalations, and outputs are inserted as color-coded inline references — called pills — using a single @ trigger, rather than typed syntax like {{argument_name}} or different syntaxes per reference type.

The editor supports Markdown formatting with an Edit/Preview toggle, a formatting toolbar with undo and redo, and a fullscreen mode. It automatically propagates renames across all prompts, highlights invalid references in red, surfaces evaluation coverage inline, and enables AI suggestion review.

Existing prompts require no changes. The editor reads and writes the same underlying format — {{argument}} for inputs and @{path} for tools, contexts, and outputs — so existing prompts open as pills automatically. Typing {{ }} directly is no longer a creation trigger; use @ to insert references going forward.

For details, refer to Prompt editor.

The following known issues affect the new Prompt editor:

As part of our ongoing cloud expansion, we're bringing complete data sovereignty to the South Korea region for Agents. This ensures your data is securely stored and replicated across multiple availability zones within South Korea, offering full control and compliance confidence. For more information, refer to the Global cloud regions page.