DataMasque joins MAS PathFin.ai for privacy-preserving AI

DataMasque has been listed as a solution provider on the Monetary Authority of Singapore's PathFin.ai programme, placing it in a library of artificial intelligence tools and implementation practices for financial institutions.

PathFin.ai is a joint initiative between MAS and the financial industry to support knowledge-sharing on AI use in finance. It brings together AI solutions and implementation examples to help institutions reduce the time needed to assess and adopt new tools.

DataMasque focuses on sensitive data discovery and masking. Its inclusion comes as banks and other financial institutions face pressure to develop and test AI models without exposing customer information or breaching privacy rules.

Data access

Financial institutions often need data that closely reflects live customer records when training or fine-tuning AI systems. This creates a challenge for organisations that must balance the need for realistic data with regulatory controls on personally identifiable information and other sensitive records.

Traditional approaches such as redaction, scrambling, and dynamic masking can reduce the usefulness of data for model training or leave some information visible. DataMasque says its system replaces sensitive data with what it describes as "synthetically identical" values that preserve format and referential integrity while keeping personal information inside a secure production environment.

The approach can also support federated learning, in which several institutions contribute to a shared model. In that setting, organisations can avoid using actual customer identities in training data while preserving the patterns a model needs to learn across institutions.

Bank project

DataMasque also pointed to a deployment at a large Asia-Pacific bank migrating an on-premises SAS credit risk system to the cloud. The bank needed to move testing and development work without creating regulatory issues linked to exposed personally identifiable information in non-production environments.

For that project, the software was deployed in an isolated air-gapped environment. According to DataMasque, the platform used metadata analysis, in-data discovery, and a locally deployed fine-tuned small language model to identify sensitive information across the bank's Hadoop storage.

It then generated masked values designed to match the original data's formats and relationships, using one-way SHA-512 salted hashing to prevent reverse engineering. The process was carried out within the bank's own environment so personally identifiable information did not leave the secure zone.

DataMasque says the work helped the bank close an open regulatory risk issue on the Chief Information Officer's risk register. It also says a third-party vendor was able to test credit risk pipelines, including risk-weighted assets and expected credit loss calculations, against masked data that maintained full referential integrity.

According to the company, this eliminated earlier pipeline failures and rework. Data refresh cycles were moving from annual updates to daily, weekly, or monthly runs, giving development and testing teams access to more current datasets.

Users needed two to three hours of training to use the platform, the company said. The proof of concept lasted about six weeks, and full deployment was completed within two months of commercial and regulatory approval.

Executive view

Grant de Leeuw commented on the difficulty many financial services organisations face when trying to move AI work from concept to implementation.

"Most AI projects in financial services stall on data access," said Grant de Leeuw, Co-Founder and Chief Executive Officer of DataMasque. "Using real customer data carries unacceptable regulatory and privacy risks, and the usual alternatives either strip out the detail a model needs to learn from or leave sensitive information exposed. DataMasque removes that trade-off: institutions get synthetically identical data inside their own environment, with the realism and referential integrity their models need. As a listed PathFin.ai solution provider, DataMasque looks forward to supporting financial institutions exploring privacy-preserving AI adoption."