Release notes

Version 4.4.1

About Release

• Release date: February 26, 2026
• Compatible platform version: Work.AI 10.3.0+

The Evan 4.4.1 patch aims to improve the AMM stability under high loads and enhance consistency in multi-step manual review and reporting workflows. The release also introduces support for Work.AI v10.3.2 Java 21 version.

• We resolved the Incorrect result size exception thrown when handling large daily entity loads, causing a portion of entities to fail during processing. With the update, Evan handles large‑scale workloads more reliably, ensuring that all entities are processed as expected even under heavy load.

• We fixed the problem where duplicate articles lost their assigned False Positive status when a manual review task was reopened. Evan now preserves the same False Positive disposition for duplicate articles throughout the task life cycle, ensuring consistent review behavior.

• We fixed the issue where AMM was unable to process subdivisions with the same country and subdivision codes. As a result, the subdivision details are handled correctly, ensuring accurate data processing and reporting.

• We added support for the Work.AI platform version v10.3.2 running exclusively on Java 21. The enhancement enables smoother upgrades and provides access to the performance and security improvements introduced in Java 21.

Version 4.4

About Release

• Release date: January 19, 2026
• Compatible platform version: Work.AI 10.3.0+

The Evan v4.4 release enhances investigation efficiency and optimizes the overall solution's performance. You can reduce noise and boost the AI Agent's accuracy with a new mechanism for managing blocked URLs, prevent redundant processing through idempotency validation, and maximize search coverage using intelligent dual-query handling for company names.

Other major updates include a redesigned two-stage disposition logic and enhanced LLM summarization and narratives. We have also expanded the AMM training dataset for short articles, streamlined task reassignment in Workspace, and resolved multiple issues to provide a smoother user experience.

New features

• Evan boasts a novel mechanism for managing URLs in the ignore list, which lets you reduce noise and improve the AI Agent's operational efficiency. To support the mechanism, the following new elements were introduced in the AMM solution:

  • The Manage Blocked URL Business Process handles enabling, disabling, or deleting related records in the amm_content_relevance_feedback Data Store.

  • The Adverse Media Monitoring (AMM) Blocked URL dashboard includes visualizations that enable you to review and modify websites to be ignored during the search.

• Evan now supports idempotency validation, enabling you to decrease the AMM workload and optimize its performance. Enabled on the Input configuration step via the Validate request uniqueness checkbox, the feature prevents reprocessing of active or completed investigations, while allowing retries for failed ones.

• To optimize the company name search, we have introduced the intelligent search with and without legal endings for the Google provider that you can switch as you configure Evan. With the search on, the AI Agent automatically executes dual queries for Company entities, searching both for the full legal name and the stripped base name, to maximize result coverage.

Improvements

• The two-stage disposition logic has been redesigned for better usability and flexibility. Now, Evan determines whether a disposition is on the first or second stage based on the group to which a task is assigned, rather than whether the investigation is opened for the first or second time. As part of this redesign, the Restricts the analyst from closing an investigation with True Positive status on the initial review? on the Human In The Loop configuration step has been renamed to Enable a two-stage review process?.

• The AMM training dataset is extended to include a representative sample of short articles (<30 words) so that the model can learn patterns better and improve prediction accuracy.

• To streamline assignment management in Workspace, we have refactored the Save and Continue flow. The updated flow allows you to re-assign a task to a different user or user group from respective drop-down lists.

• To provide comprehensive and actionable insights into the decisions of Large Language Models (LLM), we have extended LLM summarization capabilities to include False Positive dispositions. In addition, generic templates were replaced with dynamic, contextual narratives that clearly explain why an article is flagged as a non-match.

• We have redesigned the manual review workflow validation in Workspace to ensure data integrity when the Require users to disposition all articles setting is active. Now, the Needs Investigation status is automatically restricted at the investigation level once all associated articles have been dispositioned.

Bug fixes

• Fixed the display issue in the Entity information and Appendix tables. Now, both tables display "—" instead of ", ," for empty residence country fields.

• Resolved the article screen display issue where Evan failed to show the reviewer’s name for dispositioned articles lacking an adversity score.

• Fixed the issue causing the AMM Core Business Process to crash during Worker retries. Now, duplicate merge record attempts are handled gracefully: instead of throwing an exception, Evan logs a warning and continues execution.

• Resolved the performance bottleneck in the AMM Data Purge Business Process by optimizing the deletion logic to prevent near-infinite execution times.

• Fixed the issue where input validation incorrectly rejected non-ISO date formats, restoring the ability to process the dates that match custom patterns.

• Resolved the problem where the report incorrectly displayed the previous reviewer’s name after decision reapplication.