Release notes

Version 4.6.3

About Release

• Release date: July 24, 2026
• Compatible platform version: Work.AI 10.3.0+

The v4.6.3 Evan release enhances the CSV ingestion and introduces user-defined batch identifiers for improved traceability. It also resolves several configuration and processing issues to ensure more consistent behavior of the AMM solution and a smoother user experience.

Improvements

• AMM now offers more flexible and comprehensive CSV ingestion, which means less manual data preparation and better input data quality. Enhancements are as follows:

  • When you use the CSV ingestion method for External source provider, the Article URL input field is now optional.

  • CSV ingestion now supports more input fields, including Gender, Business Line, and other metadata fields available in the configuration user interface or through API. The CSV template has also been updated to reflect the newly supported fields.

• The CSV ingestion method now supports the user-defined optional batch_id parameter, enabling easier record tracking and integration with upstream systems. When the batch_id column is included in an input file, the ID value is assigned to all records in that file. Otherwise, AMM continues to automatically generate and assign a batch ID. The enhancement improves traceability and batch-level correlation while maintaining backward compatibility with existing ingestion workflows.

• You can now use an optional user-defined batch_id parameter with the CSV ingestion method, making it easier to track records and connect with upstream systems. If you include a batch_id column in your input file, that ID is assigned to all records in the file. If not, AMM defaults to generating and assigning batch IDs automatically.

Bug fixes

• Resolved the issue where the article duplicate detection threshold configured for the NLP model in the Investigation step was not retained after saving. The selected threshold value is now correctly persisted and displayed when the configuration is reopened as expected.

• Eliminated the issue where processing articles from the External source provider could intermittently fail when the Combine input and downloaded content option was selected in configuration and all filtering policies were set to Escalate. Articles are now processed and enriched reliably, preventing errors that could block escalation and require manual reprocessing.

• Fixed the problem where name matching threshold fields were not displayed when the Name matching approach option was set to Strict in the configuration user interface. The threshold fields are now visible and configurable irrespective of the selected name matching approach.

Version 4.6.2

About Release

• Release date: July 3, 2026
• Compatible platform version: Work.AI 10.3.0+

The v4.6.2 Evan release introduces summary-based fallback processing for article download failures and improves the accuracy of adverse content relevance assessment and criminal category detection.

Features

• AMM now provides summary-based fallback for article download failures, which helps preserve valuable adverse media content and reduce manual review effort. When an article cannot be downloaded, AMM automatically generates a summary or uses an available one rather than filtering out or escalating the article.

  To support the summary fallback feature, the article filtering settings in the AMM configuration interface have been updated. The Article Filtering Policy section on the Input step now uses drop-down selections instead of checkboxes, enabling you to specify whether articles with specific issue types should be filtered out, processed with a fallback summary, or escalated. The enhancement allows for more flexibility in setting the article filtering behavior across a wider range of scenarios.

Improvements

• The domain gate model now includes an optional cross-encoder to support more targeted criminal category detection. The cross-encoder adds a validation step for low-confidence predictions, allowing you to include relevant crime categories that might otherwise be missed and exclude those not supported by the article content.

• To improve focal-point identification in AMM adjudications, enhancements have been introduced to better distinguish between entities that are the actual subjects of adverse content and those that are only referenced. This reduces false-positive adverse findings and helps improve the accuracy of adjudication outcomes.

• The logic of the adverse content materiality and relevance assessment has been refined to focus on the relationship between an entity and adverse content rather than relying on keyword matches alone. This reduces irrelevant findings surfaced for review and helps focus analyst attention on more meaningful results.

Bug fixes

• Resolved the issue where template-based investigation summaries could incorrectly imply that a missing year of birth (YOB) was the reason for classifying an article as adverse. Summaries now use neutral wording for no YOB found adjudications, accurately reflecting that the article was classified as adverse and requires further review.

• Eliminated the issue where distinct articles without URLs could be incorrectly identified as duplicates due to overly restrictive article matching logic. Article uniqueness is now determined more accurately, preventing unintended reuse of historical dispositions and ensuring new articles are reviewed independently.

Version 4.6.1

About Release

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

The v4.6.1 Evan release introduces the crime detection capability:

• The Investigation configuration step now includes a dedicated Crime categories section where you can configure the following settings:

  • Select crime categories to create a unified crime list. In all, 17 different crime categories are available for selection. Each category is associated with a list of related keywords (seeds) and thresholds used to determine whether particular article contains a crime type description or not.

  • Allow Evan to escalate articles for manual review when no crimes are detected by using the Escalate article if no crimes were detected checkbox.

• The machine learning model now features dedicated crime category matching logic.

• Manual Tasks now include the Crimes section that displays configured and detected crime categories together with associated keywords. Detected crime keywords are also highlighted in article bodies.

• Final Adverse Media Monitoring reports now display detected crime types in the Entity Information section for the entire report and in the Article Details section for each article.

Version 4.6

About Release

• Release date: June 11, 2026
• Compatible platform version: Work.AI 10.3.0+

With the v4.6 Evan release, new capabilities are introduced to enable seamless reuse of historical data, enhance data retrieval, and improve ML-driven decision-making and status handling. Updates to the S3 storage configuration, article source selection, and review group setup further strengthen security, usability, and control. Other improvements include priority data propagation, better visibility of key attributes during review, and optimized text processing by the AMM model.

New features

• Evan is enhanced with the new Adverse Media Monitoring Historical Result Loader BP. Based on provided input parameters, the BP returns the results of the most recent completed investigations. The new capability reduces the manual work required to collect historical investigation data and makes it easier to reuse prior adverse media insights.

• Evan now allows conditional application of investigation statuses based on the selected article source. For Input data, you can set the AI Agent to apply only the False Positive status (default) or all incoming investigation statuses. When the Plugged connector option is selected, the investigation status is handled according to the connector’s existing logic.

• The Retrieve articles from the plugged connector checkbox has been replaced with a radio button group. You can explicitly choose between Input data (default) and Plugged connector as article sources.

• You can leverage the decision reapplication feature when Manual Tasks are disabled. Enabled in the Investigation step via the Enable Decision Reapplication? setting, the feature allows previously applied decisions to be reused automatically. This makes processing more efficient and keeps investigations running smoothly without manual work.

• New S3 storage settings have been introduced in the Output step. The settings help you define the target S3 bucket and control whether authenticated links should be stored. The new settings include:

  • The mandatory Files storage S3 bucket setting lets you explicitly choose which S3 bucket stores your uploaded files.

  • If you turn on the Store authenticated S3 links option, you can save presigned (authenticated) links instead of regular S3 object URLs.

• Evan is now integrated with the Wayback Machine for fallback article retrieval. When article content cannot be downloaded from the original link, archived versions are retrieved automatically to maintain data completeness and ensure uninterrupted investigations.

• Evan's machine learning (ML) model can now automatically disposition articles as True Positives when the confidence score is high. The capability enables high-confidence matches to be resolved without analyst intervention, speeding up investigations without affecting accuracy.

• Evan generates a complete investigation summary during processing, supporting both standard and LLM-based formats. Configurable based on your preferences, the feature reduces manual effort and delivers flexible, high-quality outputs tailored to business needs.

• You can configure Evan to use strict or partial name matching. The option lets you adjust the name matching behavior based on risk tolerance and data variability, reducing false positives and uncovering relevant matches that might otherwise be missed.

• Evan supports separate shorthand user lists for Stage 1 and Stage 2 review groups. This feature streamlines user assignment and clarifies role segmentation, improving management of multi-stage review workflows.

• The AMM AI Agent can execute Google searches through the third-party SERP API, with the option to use it as a fallback. This improves search resilience and coverage, ensuring consistent investigation quality when primary search methods are unavailable.

• Evan now integrates Browser Use to retrieve articles when content is inaccessible in the standard mode. This ensures reliable evidence collection, improves output quality, and reduces manual intervention.

• You can configure Evan to use multiple external sources instead of a single one. This capability allows for more comprehensive data retrieval, enhancing the quality and robustness of investigation results.

• The AMM ML model now detects court processes, automatically identifying court-related events in articles. The feature enables more precise classification of potential adverse media and improves investigation efficiency and consistency.

Improvements

• The _sys_mt_hit_priority field is now propagated across workflows, which helps you keep priority information for better monitoring and reporting. When included in your input data, the field is sent to Manual Tasks, saved in input and analytics Data Stores, and added to all outgoing analytics events.

• Date of birth and age are now highlighted within articles during manual review and in the final report. The enhancement helps analysts quickly identify key attributes and make informed decisions faster.

• The sentence splitting logic in the AMM ML model has been optimized for more precise text processing, resulting in higher-quality extraction and more reliable downstream analysis.

• The AMM ML model now better handles articles with a large quantity of tags. This update ensures more reliable analysis and reduces the risk of misinterpretation in complex content.

Bug fixes

• Fixed the issue where the AMM review results were ignored when the review comment started with Duplicate.

• Resolved the issue where an incorrect error code was assigned to empty articles. Now, empty articles are correctly identified and return the appropriate error code.

• Fixed the issue where investigations could be closed even when the article list was empty or not fully loaded due to task rendering errors.

• Resolved the issue preventing reliable retrieval of Akamai-protected pages. Now, such pages are correctly handled and can be successfully downloaded in the browser rendering mode.

• Fixed the issue that prevented PDF report generation in some scenarios. An HTML fallback is now used automatically to ensure uninterrupted report production.

• Resolved the issue where translation options for the External source provider were incorrectly displayed when a specific language was selected. Now, Any is set as the default language, and the Translation options section is dynamically shown or hidden based on the selected language.

• Fixed the issue where the entity type mismatch template contained the hardcoded Individual value. Now, the template correctly reflects the actual entity type.

• Resolved the issue where the daysSearchPeriod parameter was not applied by the External source provider during article filtering by publishing date. Now, the parameter is correctly enforced, ensuring more accurate and consistent retrieval of relevant articles.

• Eliminated the inconsistency where articles marked as Needs Investigation during decision reapplication failed to trigger Manual Task creation when task skipping was enabled. Now, a Manual Task is always created whenever input changes require human review.

• Fixed the issue where AMM failed to preserve the article investigation status and comments when reopening closed investigations. Now, both status and comments are retained.

• Resolved the issue where rerun investigations were not marked as COMPLETED due to status assignment being incorrectly tied to the Billing step. Now, investigation statuses are updated independently, ensuring rerun investigations are completed and reflected accurately.