Flagright

AI Forensics agents cover screening, monitoring, investigation quality assurance and governance, drafting suspicious activity reports and letting rules be written in natural language. A configurable rules engine with screening, risk scoring and case management sits beneath and would keep working without them.

Each path has a named stopping point. Agents close confirmed false positives in monitoring and true negatives in screening on their own and escalate true positives to an analyst, with Flagright reporting 99 percent escalation accuracy and 95 percent agreement with analysts. Suspicious activity report narratives are generated for human review and approval before filing, and every narrative can be edited or overridden.

The workflow builder assigns each alert, case or escalation to an agent or an analyst, agents are validated on historical data before deployment, and rule changes are proven in simulation and shadow mode before they block anyone.

Flagright tracks accuracy, hallucination rates and edit distance for its narratives over time and reports 95 percent analyst agreement and 99 percent escalation accuracy for its agents, without a method, sample or period behind those figures.

Agents are validated on historical data before deployment, each answer passes a chain of checks across four models, narratives cannot introduce facts beyond the case data, and quality assurance scores every investigated case against the team's own rubric.

Betterment, a wealth manager serving more than a million customers, halved the manual effort needed to review flagged transactions, and Banked, an account to account payments firm licensed by the Financial Conduct Authority, cut it by 83.8 percent. Sciopay went live after a seven day integration, and Flagright puts its usual integration time at five to seven days. These are customer results reported by Flagright, with no baseline period stated.

Whether one customer's data informs models serving another is not stated, and neither are the model providers the agents call or whether customer data reaches them for training. What Flagright does say is that its agents run inside its own security and data governance controls, analyst feedback refines model routing and validation logic, and narratives cannot introduce facts beyond the case data.

The privacy policy, updated September 7, 2026, reaches the platform itself. For the data an institution submits about its own customers, Flagright acts as processor under a data processing agreement, and the institution sets how long that data, the audit logs and the support records are kept. Platform user accounts are deleted or anonymized within 90 days of deactivation.

Subprocessors are described by type, cloud hosting and analytics, rather than named, and the data processing agreement itself is not published. Compliance with the General Data Protection Regulation (GDPR) and the California Consumer Privacy Act is stated, and a customer's data, logs and backups stay in the hosting region it agrees with Flagright.

ISO 27001:2022 certification and a SOC 2 Type II report are named, and stored data is encrypted under the 256 bit Advanced Encryption Standard (AES) in Federal Information Processing Standard (FIPS) 140 3 certified modules. Internal and independent specialists run penetration tests and red team exercises. Clients receive audit reports automatically through Flagright's trust portal while prospects must ask sales, and 99.99 percent global uptime is stated.

Suspicious activity and transaction reports file from the platform to FinCEN and more than 70 goAML jurisdictions, placing the product inside the regulated reporting channel. Compliance with the European Digital Operational Resilience Act is also stated.

Error rates by customer type, corridor or geography are not published, and holds, restrictions and filings fall on customers who never learn why. Behavioral rules may land harder on people with unusual but legitimate patterns, such as remittance senders, and the governance agent tracks regulatory change against policy rather than testing outcomes across groups.

Outcomes are the institution's: it configures the logic, decides how outputs are used and carries responsibility for any decision with legal or similar effect, and Flagright publishes no guarantee or remediation term. A customer of that institution who contacts Flagright about their data is referred back to the institution.

What exists is the means to catch an error before it lands: every generated narrative can be reviewed, edited and overridden, every agent action is timestamped, and audit logs export for regulatory submission.

Four models check each answer in a chain of checks and a router sends each task to the model it suits, but none of those models, their providers, the hosting provider or a subprocessor list is named. Every transaction passes through Flagright's environment, so that chain carries the customer's data.

Flagright publishes a developer hub with its application programming interface documentation and a live status page, and screening runs on each transaction before it clears. Connectors to named core banking systems or card processors are absent, so integration rests on that interface. At Betterment, a Zendesk integration feeds first line support tickets straight into case management.

The cloud provider behind Flagright's hosting regions is not named, and current material offers no hybrid or on premises route. The software runs in the cloud from more than 12 hosting regions, a customer's data stays in the region it chooses, logs and backups included, and computation runs in isolated, short lived environments.

Demo requests stand in for pricing of any kind. Flagright puts integration at five to seven days on average, which bears on total cost even without a rate card.

Each segment Flagright names has a page of its own: payment gateways, digital banks and neobanks, banks, credit unions, brokerages and trusts, crypto firms, remittance firms and gaming operators, with enterprise, mid market and small business buyers addressed separately. Insurers sit outside that list. More than 100 customers in more than 35 countries are claimed, from Betterment in wealth management to Banked in account to account payments.