Dossiers

Onboarding, screening and rule based transaction monitoring would all keep working without the AI assistants, which read customer and transaction records and work with the sources in view. The models carry triage and review work on a platform that still runs as a rules and workflow system.

AI assistants have read access to records and need an approval request for any write, analysts review matches against identity documents and source evidence before recording an outcome, and the terms leave final compliance decisions with the user. Triage thresholds and any sampling of cleared alerts go unpublished.

Each screening hit shows the attributes compared, such as a full name match beside a conflicting date of birth flagged as a likely false positive, with the source articles alongside. Precision, recall and a false positive rate never appear, nor does a validation summary.

SC Securities, NDB Securities and KOKO are among the named users on the homepage, with no alert volume, false positive reduction or investigation time reported for any of them.

Individual searches are neither logged nor retained, except when a user reports an issue, and only aggregated, anonymized usage data goes to service improvement. Whether client screening activity or analyst decisions feed its own screening data is the cross client question the terms leave unanswered.

Terms effective September 1, 2025 reach the service itself: batch screening alerts sit under the user's control and can be permanently deleted, and unexported data is deleted after a notice period of at least 30 days when an account closes. A privacy notice for the people screened, the lawful basis for compiling profiles and a subprocessor list are all unpublished.

Reasonable technical and administrative measures are all the terms commit to, with no certification, attestation or trust center named. Banks buying a platform that holds identity documents and investigation notes will ask for an attestation first.

The homepage, product and company pages name no regulator, statute or rule, and the terms are governed by Sri Lankan law. For its buyers, the Central Bank of Sri Lanka's customer due diligence rules would be the natural reference.

Elimination is conservative: a match is cleared only where an attribute positively contradicts identity, never on a missing attribute or country alone. South Asian names produce more transliteration variants and false matches than the Latin script names most matching systems were tuned on, and no accuracy by script or region is published.

Under the terms the service is provided as is, screening accuracy is not warranted, users must verify every result, regulatory penalties are excluded and liability is capped at twelve months of fees. Nothing in the terms gives the person behind a false hit a way to contest it or correct a profile.

Screening runs against its own Dossiers Cosmos data or a supported provider, but no sanctions, PEP or adverse media supplier is named, nor any model provider behind the assistants. Sanctions and PEP data usually comes from a small number of global licensors, so which one sits underneath matters to a buyer comparing incumbents.

Records arrive by API, the platform works alongside existing customer and screening systems, and it can connect to payment systems to block transactions. Yet no core banking or case management system is named, and no developer documentation is linked.

Hosting model, region and residency all go unstated, though several regulators in its markets expect customer data to stay onshore. Run by a Singapore company from Colombo for South Asian institutions, the platform meets that question directly.

Price, tiers and billing basis are all unpublished, and engagement runs through a demo booking. In South Asian markets, where the alternative is often a global screening vendor priced for a different economy, the cost position would differentiate it and is not stated.

Regulated financial firms in Sri Lanka and the region are the market, with securities brokers, wealth managers and a payments company among the named users. Material is organized by module rather than institution type, and no bank appears among the named users.