Bretton AI
Bretton AI sells AI agents that do the back office work of banks and fintechs, led by anti money laundering, know your customer and sanctions reviews. It renamed itself from Greenlite in February 2026, when it raised a $75 million Series B. The platform offers more than 30 skill templates and more than 180 data sources, a Builder for writing new agents in plain language and a Workbench where analysts work cases beside the agents. The managed service replaces an outsourced compliance team. Named onshore investigators take the judgment calls, every case is checked against the client's quality scorecard, and fees follow completed cases.
Banks can run it on Bretton AI's platform, on dedicated infrastructure or in their own cloud account. Kaufman Rossin independently validated its compliance agents in October 2026. First Internet Bank, Grasshopper Bank, MVB Bank and Payward, the parent of Kraken, are named customers.
Capability Axes
Capability grades
15 of 15 axes rated · 13 graded A or B
Agents complete financial crime compliance work that outsourced analyst teams used to do, drawing on more than 30 prebuilt skill templates, and a Builder lets a team define new agents in natural language. What the agents replace, a human workforce, is all that would be left.
Agents gather evidence, apply the institution's policy and draft the full case for its team to approve. Lower risk decisions can close automatically, while high risk workflows run in human review modes. Confidence bands show how strongly the evidence supports each screening outcome, though no routing threshold is published. Workflows link agents into one investigation, with an analyst confirming each hand off.
In the managed service, named onshore investigators take the judgment calls and report work. A trained analyst reviews every output before it reaches the client, and every case is checked against the client's own quality scorecard.
Kaufman Rossin, an accounting and advisory firm, completed an independent model validation of the know your customer, anti money laundering and sanctions agents in October 2026. It covered accuracy and stability, explainability, change management, monitoring and governance, and no findings are published. Before material changes ship, agents are tested against synthetic and adversarial cases and benchmarked against human validated production cases.
The Bretton Evaluator then checks production runs after launch. Every decision is logged with its reasoning, evidence and model version. Bretton AI says the work is built to federal banking guidelines but names no supervisory instrument. SR 26 2, which replaced SR 11 7 in April 2026, puts generative and agentic AI outside its scope.
First Internet Bank cut compliance review time for new customers from 10 hours to under an hour and saves 24,300 hours a year. Payward, the parent of Kraken, saved $3.8 million in its first year and more than 52,000 hours of manual work across more than 21,000 cases since integration, and its chief compliance officer, CJ Rinaldi, is quoted. Grasshopper Bank cut enhanced due diligence time by 70 percent, and Sling, Fluz and HitPay are named with results of their own. The figures come from Bretton AI's case studies, with the period stated for First Internet Bank and Payward.
Bretton AI acts as a processor for its customers. It holds zero data retention agreements with every model provider it uses, and inference is processed in memory and never stored. In a bank's own cloud account, customer data stays there and only usage telemetry returns to Bretton AI. The trust center commits to deleting data once business or regulatory need ends. The platform page says every case makes the next one sharper, and nothing states whether one client's cases improve the agents serving others.
The privacy policy, effective January 28, 2026, covers website visitors and users of the services and names only Stripe, for payments. It does not address the client data processed for banks, though a June 2026 partner post says Bretton AI acts as a processor for its customers. The published terms tell users not to use the services in a way that would violate the Gramm Leach Bliley Act. The data management policy sits behind a trust center request. Identity documents, financial statements and adverse media on the people under review pass through the agents.
A Wolfia powered trust center lists a SOC 2 Type II report for January 2025 to January 2026 and fifteen security policies, all released on request. It also describes annual penetration tests, quarterly external scans and AES 256 encryption at rest. The listed report period closed in January 2026, so currency depends on a newer report or bridge letter.
Bretton AI sells software and services and claims no license, which is expected for a supplier. Its clients include banks supervised by the OCC, the FDIC and the Federal Reserve, and each client keeps control of its program, decisions and filings. A September 2026 blog post calls the platform regulator approved but names no regulator and no approval.
An August 2026 fairness analysis of the Research Agent publishes its method and results. About 1,000 synthetic people and companies from 129 countries were checked field by field against human labeled answers. For 44 people with names in Cyrillic, Arabic, Chinese characters or Vietnamese, recall was 100 percent, with three wrong region values traced to a weak source. Precision was 98.6 percent for country and 94.2 percent for region. The figures are Bretton AI's own and cover one agent on synthetic data. No breakdown is published for the screening alert agents, where uneven name matching lands on real customers.
Every agent decision is logged with its reasoning and evidence, which gives an institution a trail to challenge an outcome. The published terms supply the services as is, with no warranty of accuracy. They cap liability at six months of fees and have the customer indemnify Bretton AI, so the loss sits with the customer. No accuracy guarantee is published, and nothing routes a customer whose alert was wrongly handled.
Bretton AI says it holds zero data retention agreements with every model provider it uses, so inference requests are processed in memory and never stored. In a bank's own cloud, inference can run through an approved Amazon Bedrock endpoint. The model providers are not named, the trust center lists no subprocessors and the privacy policy names only Stripe.
Logos for nCino, NICE Actimize, ServiceNow, Oracle, FICO and ICE sit under a line saying the agents work inside the systems a team already runs. Native connectors are claimed for case management, screening and core systems, with no core system named. Unit21 and Databricks integrations launched in February 2026, and Equifax and LexisNexis Risk Solutions data can feed investigations directly. No developer documentation or marketplace listing that would confirm depth is linked from the site.
Banks choose between Bretton AI's shared platform, dedicated single tenant infrastructure and their own cloud account. In its own account the bank governs compute, storage, keys and monitoring, customer data stays there, and only usage telemetry returns to Bretton AI. Agents can call only approved models in the approved region, and a fully disconnected setup runs without web research or hosted frontier models. The terms say the hosted service runs in the United States and the United Kingdom.
The managed service is billed on consumption from the first day, with fees tied to completed cases rather than analyst hours. Bretton AI claims half the price per case of a typical outsourcer and quotes a cost per case after reading a prospect's current outsourcing contract. No rate, tier or platform price is published.
Banks supervised by the OCC, the FDIC and the Federal Reserve are the stated buyers. Payments companies, crypto exchanges and brokerages also appear in the case studies. The work covers due diligence, sanctions and transaction monitoring, and use case pages extend to SBA loan review, HMDA data checks, underwriting and vendor risk. Credit unions and insurers get no dedicated material, and the focus is the United States.
Product updates
Material product, regulatory, evidence and commercial changes at Bretton AI, each verified against a live source and tagged to the capability axis it bears on. Funding rounds and awards are not product changes and are not logged.
Bretton changed how its AI drafts quality control scorecards, so the draft now reviews the whole case: the narrative, findings, case inputs, transactions and the agent's investigation record. It also draws on the team's past reviews of the same scorecard. The draft lists specific problems first, each backed by a quote from the case, then answers each scorecard question from them, and past reviews carry over when a scorecard is renamed or versioned.
Bretton changed the related entity check in onboarding screening for cases where a screening provider's match profile merges different organizations with similar names. The check now works out which organization is the customer, and returns no result when it cannot tell rather than reporting a false mismatch. The change applies automatically, with nothing to switch on.
Bretton improved Agent Builder so teams can build, edit and validate the structured results and text reports their agents return, with guided field editing, setup from a sample, and validation before a change is saved. Changing a downstream output format previously meant manual configuration and test runs.
Compared With
Most editorial comparisons pair two vendors the index assesses as direct competitors for the same buyer. Some pair vendors that are adjacent rather than rival, where the useful question is where one ends and the other begins. Each carries a verdict, the buyer conditions that favor each vendor, and a graded side by side.
Alternatives to Bretton AI
Other products that do the same kind of work as Bretton AI, and what each one does.
Tookitaki unifies anti money laundering monitoring, fraud prevention, screening, onboarding risk, customer risk scoring and case management in FinCense, a platform for banks, digital banks, digital wallets, remittance firms and payment institutions across Asia Pacific.
Lucinity runs financial crime investigations for banks, payment firms and fintechs in two ways.
Flagright sells a financial crime compliance platform to payment processors, digital banks, banks, credit unions, brokerages, crypto and remittance firms, covering real time transaction monitoring, dynamic customer risk scoring, sanctions and watchlist screening, case management and…
Quantexa, a London company founded in 2016, sells a Decision Intelligence Platform that resolves scattered records about people and companies into one view and maps the network around each one.
Consilient sells anti money laundering and counter terrorist financing models built with federated learning, so banks, regulators and financial intelligence units can learn from each other's behavioral patterns without pooling their data.
Napier AI is a London company, founded in 2015 and majority owned by Marlin Equity Partners since February 2025, selling anti money laundering software to banks, payment firms, wealth and asset managers and insurers.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | Data Protection Terms | Implementation | Source |
|---|---|---|---|---|
|
Not published. The managed service is charged by completed case, and Bretton AI claims half the price per case of a typical outsourcer.
|
The managed service is billed on consumption from the first day, with fees tied to completed work rather than analyst hours. Bretton AI quotes a cost per case after reviewing a prospect's current outsourcing contract. No basis of charge is published for the platform, its templates, the Builder or the Workbench. | Data protection terms are not tiered. | No implementation fee is published. Bretton AI says there is no migration project and the first workflow goes live within 60 days of contract. | Vendor Published |
Bretton AI charges for its managed service by completed case, billed on consumption from the first day, and claims half the price per case of a typical outsourcer. It publishes that billing basis, not a rate.