The Bretton Ledger

The latest from product teams at Bretton AI.

Agents and WorkflowsNew

Structured outputs for agents

What changed

An agent can now produce a structured output alongside its narrative. You define the shape you want, such as a JSON object, booleans, arrays, strings, or a text report, in the agent's output configuration, and a final step conforms each run's findings and narrative to that shape.

Why it matters

Consuming agent results in another system meant parsing free text or prompt-engineering the format.

Customer impact

Integrate agent results into your own systems in whatever format you need, which speeds onboarding.

InvestigationsImproved

Show or hide citations in narratives and exports

What changed

Citations can now be shown or hidden on the narrative page and in exported PDFs and workbooks, in bulk from the investigations and agents page or one case at a time.

Why it matters

Analysts move finished narratives into case-management systems and sometimes need the text without inline citations.

Customer impact

Choose whether citations appear in what you export or copy, without reformatting by hand.

Controls and TrustNew

QC scorecards for case review

What changed

Reviewers can grade analyst work on a case against a scorecard. Scorecards are configurable per agent and per organization, every template change is versioned, and a reporting page shows pass rate by month and by analyst.

Why it matters

Quality control was happening off-platform in spreadsheets, with no coverage tracking.

Customer impact

QC lives on the case itself, with aggregated reporting teams can act on.

InvestigationsNew

Workbench as an assistant on agent cases

What changed

Workbench is now available as an assistant on completed agent cases, with access to the case data, the organization's skills, artifacts, and direct narrative updates.

Why it matters

Analysts had to leave the case or move findings by hand to dig deeper.

Customer impact

Investigate, reason, and update a case narrative without leaving the case.

Agents and WorkflowsNew

Rerun a case from any stage

What changed

You can rerun an agent case from transaction analysis, entity enrichment, skills, or narrative generation, update the entities, affiliates, documents, and context, and choose the agent version for the rerun.

Why it matters

Changing an input or configuration meant creating a whole new case and losing history.

Customer impact

Rerun only the stages you need, and keep the audit history intact.

Agents and WorkflowsNew

Disposition-based narrative templates

What changed

An agent can now choose a narrative template based on the case's final disposition, from a list you configure, rather than using one narrative for every outcome.

Why it matters

Different outcomes need different write-ups.

Customer impact

Narratives match the disposition automatically. Off by default, configurable in Builder.

Screening and MonitoringNew

Transactions view for transaction monitoring

What changed

Transaction monitoring now has a native transactions view. Alerted transactions surface at the top with aggregated stats, and you can filter, sort, switch between money-in and money-out, and aggregate by counterparty, all in product.

Why it matters

Teams were exporting transactions to spreadsheets to analyze them, which was slow and hard to audit.

Customer impact

Review and analyze monitored transactions without leaving Bretton or building a spreadsheet.

Details and FAQ
  • — Do I still need to download a CSV? No, not necessarily. The view replaces the export-to-spreadsheet workflow. You can still download an xlsx workbook that contains case output.
  • — Can I focus on just the alerted transactions? Yes. Toggling alerts filters both the chart and the table.
InvestigationsImproved

Narrative editing in Workbench

What changed

Investigation narratives are now fully editable as formatted text. Citations render as clickable footnotes, tables are editable, and citations renumber automatically as you add or remove them.

Why it matters

Editing used to be piecemeal and easy to get out of sync.

Customer impact

Analysts edit and finalize a narrative in one place, with citations and an audit log that stay correct.

InvestigationsImproved

Case metadata in the attachment view

What changed

Customer metadata now appears directly in the case attachment view, with a download option.

Why it matters

Analysts had to look this up separately.

Customer impact

Less context switching during a review.

IntegrationsImproved

Bulk alert ingestion, expanded

What changed

Bulk alert ingestion now covers both screening and payment-screening alerts through the API, accepts CSV and JSONL, and supports encrypted delivery and secure file transfer for enterprise pipelines. This supports batches on the order of tens of thousands of alerts.

Why it matters

Enterprise programs move alerts in large, sensitive batches and need secure, reliable ingestion.

Customer impact

Move very large alert volumes into Bretton securely, with errors returned for correction.

Major launchPlatform and ReliabilityNew

Introducing the Bretton AI-native managed service

Read announcement

What changed

A new operating model that combines AI-native execution with experienced subject-matter experts, human oversight, quality control, and program development.

Why it matters

Financial institutions need to adopt AI without taking on the capital investment and operational burden of building a new program alone.

Customer impact

Bretton's forward deployed engineers and industry experts build, train, tune, and operate AI-native programs around customer outcomes.

Major launchPlatform and ReliabilityNew

The Bretton AI Platform launches

Read announcement

What changed

Build, run, and govern AI agents across the financial back office, with managed deployment and audit-ready output from day one.

Why it matters

The banking back office no longer has to grow every time the bank does.

Customer impact

Financial institutions can transform the back office with agent-first operations built around their data and policies.

Agents and WorkflowsNew

Scope skills by entity role

What changed

In Builder you can now choose where each skill runs, on the subject, on affiliates, or on counterparties, and give role-specific instructions for each.

Why it matters

One generic skill prompt for every entity type produced noisy findings.

Customer impact

More precise multi-entity investigations.

Controls and TrustNew

Audit log for agent runs

What changed

Agent runs now have a persistent, chronological audit log covering lifecycle events, skill execution, tool calls and results, and the reasoning phases behind the output, shown in a panel on the case.

Why it matters

Teams need a clear record of what an agent did and how it reached its output.

Customer impact

Agent work is easier to review, trust, and explain.

Agents and WorkflowsNew

Build transaction monitoring agents from a CSV

What changed

You can now build a transaction monitoring agent in Builder from a CSV. The mapper proposes the transaction schema, you review the fields and value mappings, and the agent ingests, analyzes, and enriches the transactions at runtime.

Why it matters

Standing up a transaction monitoring agent for a new data format took custom engineering.

Customer impact

Stand up transaction monitoring agents faster, with the agent working from the full transaction data.

Screening and MonitoringImproved

Better export and filters for screening alerts

What changed

The screening-alerts CSV export now matches what you see in the table, and you can apply multiple filters, including created and updated dates.

Why it matters

Exports did not line up with the on-screen view.

Customer impact

Export exactly what you are looking at, filtered how you need it.

Agents and WorkflowsNew

Builder launched

What changed

Builder is live. Teams can create, configure, test, and publish agents for workflows like enhanced due diligence and transaction monitoring, without engineering. Launched as Agent Builder.

Why it matters

Creating and updating agents used to require engineers.

Customer impact

Business and operations teams stand up and maintain their own agents. Use cases expanded beyond financial crime into compliance, fraud, risk, and other banking operations.

Screening and MonitoringNew

More screening match tags

What changed

New match tags give finer control over entity matching: nationality, citizenship, registration, country of incorporation, gender, and street address.

Why it matters

Programs need granular control over what counts as a match.

Customer impact

Tune matching and dispositions to your risk rules, from the tag settings.

InvestigationsNew

Export investigations to PDF

What changed

Investigation artifacts can be exported to a shareable, high-quality PDF, with sources included.

Why it matters

Moving investigation output into compliance workflows meant manual copy and paste, with formatting drift.

Customer impact

Download a clean, sourced report in one step.

InvestigationsNew

Bring your own knowledge base

What changed

Organizations can upload their own documents, policies, procedures, and guidance into a per-organization knowledge base that Workbench searches and reads during an investigation.

Why it matters

Anything customer-specific, like an internal PEP-escalation policy, used to live outside the tool in the analyst's head.

Customer impact

Drop your compliance documents in once, and investigations use them automatically.

InvestigationsImproved

Template variables in Platform

What changed

Prompt templates now support variables, such as a business name or website, that you fill in when you use the template.

Why it matters

The same investigation setup applies to many different entities.

Customer impact

Reuse a setup across cases without rewriting it each time.

Controls and TrustNew

Research and enrichment quality, measured against ground truth

What changed

A system now scores how accurately Bretton's research and enrichment perform against verified ground truth, and folds each verified result back into a growing reference set. This includes precision, recall, and F1.

Why it matters

Research quality has to be measured, not assumed, especially where the work is open-ended.

Customer impact

Enrichment and research get measurably more accurate over time as well as more complete, better-grounded entity profiles.

InvestigationsNew

Audit log for Workbench investigations

What changed

A sidebar audit log shows exactly what an investigation did behind the scenes: every response, every tool call, and the parameters it used. The enrichment display in screening alerts now shows, for every field, both the source and the reasoning that led to it, not just a summary.

Why it matters

Users could see loading states but not an in-depth record of the investigation.

Customer impact

A transparent, reviewable record of every investigation.

InvestigationsNew

Attach documents in an investigation

What changed

You can attach PDFs, spreadsheets, and other documents directly in an investigation by drag-and-drop or file picker, and ask the agent to analyze them.

Why it matters

Analysts had to copy and paste document content or switch tools.

Customer impact

Upload shareholder registers, incorporation documents, or statements and analyze them in place.

Screening and MonitoringImproved

Faster transaction monitoring

What changed

The transaction monitoring pipeline was rebuilt to research counterparties inline instead of spinning up full due-diligence cases, to store transactions as structured data, and to let teams edit narratives directly in the interface.

Why it matters

The older pipeline was slow and costly on high-volume cases and locked the narrative to a fixed template.

Customer impact

Much faster, lower-cost transaction monitoring, with editable output.

InvestigationsNew

Reusable prompt templates in Workbench

What changed

Workbench now offers default prompt templates and lets analysts save their own.

Why it matters

Analysts repeat similar investigations often.

Customer impact

Faster, more consistent starts to common investigations.

Screening and MonitoringImproved

Consistent country and jurisdiction matching

What changed

Country and jurisdiction names now resolve to a single canonical standard across the platform, so variants like Ivory Coast and Cote d'Ivoire, or Myanmar and Burma, are treated as the same place.

Why it matters

Conflicting country data will cause avoidable mismatches.

Customer impact

More accurate name and jurisdiction matching.

Screening and MonitoringNew

Chinese name matching for payment screening

What changed

Payment screening can now match Chinese names across simplified and traditional variants, pinyin romanizations, cross-script comparisons, and company-suffix variations.

Why it matters

These matches previously all fell into ‘needs investigation.’

Customer impact

Higher automation and less manual review for programs screening Chinese entities.

Screening and MonitoringNew

Payment Screening

What changed

Payment screening is now a first-class product: screen the parties on a payment against sanctions and watchlists, review hits in a purpose-built, hit-by-hit interface with bulk actions, parse SWIFT MT and ISO 20022 payment messages, and take alerts from third-party screening providers through a direct integration.

Why it matters

Payment screening is a false-positive-heavy workflow where slow or unclear review can hold up legitimate funds.

Customer impact

Faster, clearer payment-screening review, in-product and by API.

Screening and MonitoringImproved

Screening-alert citations

What changed

Screening-alert citations now render as complete, working links.

Why it matters

Citations were sometimes missing or malformed, which undercut auditability.

Customer impact

Analysts can trust the citations and click through to the source.

Controls and TrustNew

Screening Alerts Configuration UI

What changed

Teams can now manage screening tags, rules, and alert dispositions directly in the platform. Admins update tag configurations and the rules that determine how alerts are dispositioned; other users have read-only access. Configurations can also be set per jurisdiction or per alert queue.

Why it matters

Changing screening behavior used to require a manual request.

Customer impact

Compliance teams tune their own rules and tags, with clear limits on who can edit.

Details and FAQ
  • — What can I configure? View all screening tags, adjust thresholds, turn optional tags on or off, and edit rules to drive more false-positive or true-positive dispositions.
  • — Who can edit? Admins edit. Everyone else reviews.
Controls and TrustNew

PEP Risk Configuration

What changed

Organizations can set how high-risk politically exposed person matches are classified, and can auto-close lower-tier PEP classifications while holding the highest tier for manual review.

Why it matters

PEP risk appetite varies by institution, and analyst attention should go to the highest-risk cases.

Customer impact

More control over PEP decisioning, configured from the Screening Alerts configuration UI. Auto-close is opt-in and configurable.

Screening and MonitoringImproved

Improved Document Recency Checks

What changed

Document recency and expiration checks now use an explicit reference date, so valid documents are no longer wrongly flagged as expired or future-dated.

Why it matters

Ambiguity about the current date was producing false flags.

Customer impact

Fewer false positives and more reliable KYC and KYB onboarding assessments.

Screening and MonitoringNew

Local Language Adverse Media Search

What changed

Adverse media can now be searched in an entity's local language, with results translated back to English for review.

Why it matters

Critical adverse media often exists only in local-language sources.

Customer impact

Broader regional coverage and no manual translation, so important local information is not missed.

Screening and MonitoringNew

Multiple Website Risk Assessments

What changed

Enhanced due diligence can now assess more than one website for a single company entity.

Why it matters

Many businesses operate several websites, and assessing only one misses risk.

Customer impact

More complete due diligence for multi-site businesses, with less manual effort.

IntegrationsNew

Databricks Delta Share Integration

What changed

Bretton supports Databricks via Delta Sharing, so organizations can securely connect their Databricks data warehouse and bring relevant data into Bretton.

Why it matters

Much of the context that improves compliance work lives in the data warehouse.

Customer impact

Databricks users connect their data securely and give Bretton more to work with.

IntegrationsNew

Unit 21 Integration for Transaction Monitoring

What changed

Bretton integrates directly with Unit 21 for alerts and case management, retrieving transaction alerts in real time and reviewing historical alert outcomes and past SARs.

Why it matters

Clients on Unit 21 needed their alerts and case history to flow into Bretton.

Customer impact

Teams review and act on Unit 21 cases as soon as they are created, with historical context included.

Screening and MonitoringImproved

Enhanced Transaction Monitoring Analysis

What changed

Transaction monitoring now delivers deeper counterparty insight and clearer alert narratives, with due diligence run automatically on top counterparties and structured reports that carry the full context of an alert.

Why it matters

Analysts were assembling this context by hand.

Customer impact

Faster, better-informed decisions, with the relevant context already in the report.

Details and FAQ
  • — Generates Enhanced Due Diligence cases for the top counterparties in an alert, with no manual case creation.
  • — Enriches counterparty entities with business information before due diligence begins.
  • — Detects typologies such as structuring, rapid movement, and value deviation at a finer grain.
  • — Reads uploaded KYC documents to pull entity details, including dates of birth, names, directors, and shareholders, into the customer profile.
  • — Factors an entity's website and industry-risk context into product-line analysis.
  • — Produces one structured narrative that brings transaction detail, counterparty analysis, and due diligence findings together.
Screening and MonitoringNew

Periodic and triggered reviews

What changed

Enhanced due diligence can now run as a periodic review or a triggered review, carrying the cause and prior context of the re-review into the case and its narrative.

Why it matters

Reviewing an already-onboarded customer is different from onboarding one, and that context matters.

Customer impact

Run periodic and event-triggered re-reviews as a repeatable playbook, not a fresh case each time.

Screening and MonitoringNew

Aged Adverse Media Tag

What changed

A new Aged Adverse Media tag reviews the adverse media behind an alert, finds the most recent adverse event, and flags the alert as a likely false positive when the activity is older than a threshold you set.

Why it matters

Outdated adverse media generates review work that rarely changes the outcome.

Customer impact

Less time on stale alerts, more on higher-risk ones. The threshold is configurable, and the tag is turned on from the Screening Alerts configuration UI.

Screening and MonitoringImproved

Employment Match Tag Enhancements

What changed

Employment matching now understands the meaning of employer names, so minor differences like ‘Apple’ and ‘Apple Inc.’ are recognized as the same employer.

Why it matters

Small naming differences were causing missed matches.

Customer impact

More accurate alerting and fewer false negatives, so teams decide faster and with more confidence.

Screening and MonitoringImproved

Address Verification improvements

What changed

Address verification now includes a registered-agent check alongside the existing checks for PO boxes and virtual offices.

Why it matters

A registered agent at an address can signal address clustering or shell-company risk and raises questions about whether a company really operates there.

Customer impact

More rigorous address analysis and more confidence in location-risk assessments.

Screening and MonitoringImproved

Adverse Media improvements

What changed

Adverse media search now works more like an analyst doing research, using shorter, more targeted queries to find more relevant results.

Why it matters

The older approach missed findings a human researcher would have caught.

Customer impact

Broader, more relevant adverse media coverage.

Details and FAQ
  • — Configurable keywords. Choose up to 20 search terms, or use the default set.
  • — Cleaner results. Searches run without carrying over bias from prior sessions.
  • — Smarter queries. The search can use details like date of birth or location to improve accuracy.
Screening and MonitoringNew

New Transaction Analysis Assessment

What changed

Enhanced due diligence cases and transaction monitoring alerts now include a deeper transaction analysis that breaks activity down by debits and credits, purchases and sales, and top counterparty, and flags patterns such as rapid funds movement, structuring, and unusual changes in frequency, value, or source and use of funds.

Why it matters

Line-by-line transaction review is slow and easy to lose the thread of.

Customer impact

Teams see the high-level trends and the flagged patterns without reading every transaction.

Screening and MonitoringNew

RFI Analysis

What changed

RFI Analysis reviews customer documentation, including identity documents, bank statements, and payslips, against your RFI criteria, and checks each document for identity, location, and income discrepancies.

Why it matters

Reviewing request-for-information responses by hand is slow and inconsistent.

Customer impact

Faster, more consistent RFI review, with discrepancies surfaced automatically.

IntegrationsNew

Sumsub Integration

What changed

A two-way integration with Sumsub pulls new alerts into Bretton, works them, and sends the results back, with an option to close false-positive evaluations automatically.

Why it matters

Sumsub users needed their alerts handled without leaving their existing setup.

Customer impact

Alerts flow both ways automatically, and false positives can close on their own.

Grow the bank. Not the back office.

Contact Us