New Features:

• Single-sign-on (SSO) authentication with customer IdPs over OpenId Connect (OIDC)

Product Enhancements:

• The toxicity classifier model was retrained with an upgraded harmful request dataset with better false examples including innocent requests
• All PII configurations are exposed in the UI for creating, editing, and viewing the task rule
• The document selection of the Chat UI will now persist for the entirety of a user’s session

Bug Fixes:

• Fixed an issue where refreshing while on the task page in the shield admin UI would bring the user to the inference deep dive page
• In version 2.0.293, the prompt injection rule was not able to evaluate prompts larger than 512 tokens

Product Enhancements:

• Improved the prompt injections rule with a focus on jailbreaking and instruction manipulation
• The latency metric in milliseconds is captured for each Shield rule execution

API Change:

• Eliminated the regex field from the RegexConfig schema that was deprecated and replaced by regex_patterns in v2.0.190 (January 8, 2024 release)

Product Enhancements:

• Upgrades on the toxicity rule
  • The toxicity rule has been expanded to check for additional categories including harmful/illegal requests
  • The new classifier is trained on more recent and bigger data that has better scores on our benchmarks than our previous classifier
  • The regex component of the toxicity rule received more advanced capabilities to flag profanity and attempted obfuscation of it
• New PII Rule configurations
  • confidence_threshold has been added to ignore PII entity recognitions with confidence scores below the specified threshold
  • disabled_pii_entities is added to disable a certain set of PII entities from getting evaluated
  • allow_list is added for bypassing validations of certain texts
• An informative and helpful error message is returned when the provided config field is invalid during the rule creation process

Bug Fix:

• Adjusted the presentation of PII validation details of the prompts in Chat UI so they’re fully legible

Bug Fix: The default rules were previously enforced on up to 10 tasks. The default rules now work on all active tasks.

Product Enhancements:

• Introducing the Shield Admin UI including 1) a log of inferences and 2) the ability to create and manage rules by task

Bug Fixes:

• Improved handling of authenticated session expirations in the UI
• Improved scrolling through prompts and responses in the Chat UI

• Optimized inferences query endpoint; reducing latency by 90%
• Added page size limit of 250 items to all paginated endpoints
• Fixed toxicity configuration not returned when retrieving rules

• The RegexRule now supports multiple regex patterns per rule. When creating a new RegexRule, please use the new regex_patterns field instead of the now deprecated regex field in the JSON request.
• The inference query endpoint now supports the capability to filter by task IDs, providing more flexibility in retrieving specific data
• Added the ability to search tasks and rules. While the "Get All" endpoints for tasks and rules are still available, they are now deprecated in favor of the new search endpoints. This enhancement allows for more efficient retrieval of tasks and rules.
• Introduced the functionality to slice and dice token usage by task IDs and rule types

Minor bug fixes

• Regex and keyword rules check now returns span/detected text
• The PII rule now returns confidence scores
• Added the capability to filter processed inferences on the query endpoint by rule type, rule status, and validation result status
• Enhanced the consistency of hallucination check

• Added Toxicity Rule check
• Added user-generated API keys workflow. Admin users can now create and rotate API keys using a master key
• Made modifications to improve the precision of the Hallucination rule