You can see the latest product updates for all of Google Cloud on the Google Cloud page, browse and filter all release notes in the Google Cloud console, or programmatically access release notes in BigQuery.
November 18, 2024
Two major engine versions within the v4 tuning version are no longer used by customers and are deprecated as of today. We recommend customers use the most recent engine versions instead. Deprecation overrides the support timeline for all minor versions within these major engine versions.
October 31, 2024
A new major engine version is available for Retail and Commercial lines of business, within the v4 tuning version. These engine versions:
Introduce a new feature area within the unusual-counterparty-activity feature family focused on surfacing suspicious parties through their inbound and outbound transactions with exited parties.
Apply a new data validation to ensure there are no periods in the required time range without any valid entries in the Party, Transaction, or AccountPartyLink table.
The retail engine version also has more reliable tuning performance, in particular for small datasets. This improvement was already present in commercial engine versions.
October 14, 2024
The API is now available in the australia-southeast1 region. For more information on supported regions, see AML AI locations.
October 01, 2024
A new major engine version is now available for Retail and Commercial lines of business, within the v4 tuning version. This includes:
- New recommended field
counterparty_account.region_codeadded to the Transaction table. - The new engine version uses this field to account for risks associated with the region of the counterparty account.
September 16, 2024
A new major engine version is now available for Retail and Commercial lines of business, within the v4 tuning version. This includes:
- Reduction of the total requirement for Transaction and Account data from 41 to 30 months
- Performance improvements across several feature families, focusing on more recent high risk activity
- Adjustment to the calculation of the PartyRecall metric in the rare corner case when many customers have the same prediction score and it's not possible to yield exactly
partyInvestigationsPerPeriodpositive predictions - Uses the latest FATF high risk geos, published in Jan 2024 (High-Risk Jurisdictions subject to a Call for Action and Jurisdictions under Increased Monitoring)
July 11, 2024
A new major engine version is now available for Retail and Commercial lines of business, within the v4 tuning version. This includes:
- Additional data validation errors with more granular checks and corresponding actionable error messages
- Improved accuracy and better descriptions for existing data validation checks
- A fix for processing of alert events in the Risk Case Event table
- Improved reliability of training, prediction, and backtesting operations for very large datasets (greater than 20 million parties)
- Reduction in the time taken for tuning when creating an engine config
April 25, 2024
A new major engine version is now available for Retail and Commercial lines of business, within the v4 tuning version. This includes:
- More sensitive skew metrics for better model and data quality monitoring
- A bugfix for risk score threshold estimation used in recall metrics in AML AI resource metadata
March 15, 2024
Improved the party de-registration process. You can now remove parties without prediction intent (that is, those parties not included in a create prediction results request) within a 45-day window following registration.
March 13, 2024
Released a new v4 engine versions for the commercial line of business, with more reliable tuning performance, in particular for small datasets.
March 12, 2024
Added a new metric to AML AI, providing insight into the importance of each feature family to an AML AI Model. This metric is available in new v4 engine versions. It allows you to:
- Act on model monitoring outputs in the context of their importance to a model
- Check the contribution of your Party Supplementary Data to a model
March 04, 2024
AML AI has improved handling of supplementary risk indicators included in your datasets. This includes:
- Release of new engine versions within both v003 and v004, improving usability of party supplementary data. You can now use letters, numbers, and underscores for the party supplementary data ID.
- Addition of new data validations for party supplementary data IDs.
Save time and cost when adopting new EngineVersions:
- For new engine versions, including versions in v003 and v004, you can now inherit hyperparameters from an existing engine config instead of re-tuning. This leads to quicker creation, and there are no additional costs for tuning.
- All of your existing engine versions can be used as a source for inheriting hyperparameters.
- See Configure an Engine to find out more about how this works.
February 28, 2024
Added a new engine version page so you can keep track of the latest engine version releases.
January 22, 2024
Added a quickstart guide and a sample dataset to use with it. You can use these together for end-to-end functional testing of the AML AI API prior to, or in parallel to, moving sensitive customer data to Google Cloud.
December 01, 2023
V4 engine versions for retail and commercial lines of business are now available. These engine versions extend support to datasets with up to 130 million parties and 2+ years of transactions. The retail engine versions include new KYC feature families including occupation, tenure, and assets for improved risk detection performance.
The following fields are now included in the AML input schema:
occupationcivil_status_codeeducation_level_codeassets_value_range
November 15, 2023
V3 engine versions for retail and commercial lines of business are now available. These engine versions improve the labeling methodology and address bugs in earlier engine versions, improving both reliability and performance.
June 29, 2023
AML AI is generally available with release version v1.
The API supports the following capabilities:
- Model tuning through
engineConfigresources - Backtesting and prediction using a model
- Exporting metadata from an engine config, model, backtest, or prediction resource