Silent Eight

Iris 7 is policy bound agentic AI that executes sanctions, anti money laundering and customer due diligence decisions through named agents for screening, transaction monitoring and adverse media. Without the models, no screening adjudication would be left to sell.

Agents execute decisions rather than only recommending them, bound by the institution's policies, with ownership and escalation kept by the institution. The point where an agent stops and escalates, and how often closed alerts are sampled, are not published.

Each outcome comes with its applied reasoning and supporting evidence, and quality assurance is a named control. Up to 99 percent precision is stated without a method, test set or recall figure, and model risk management is mentioned with no supervisory instrument named.

HSBC's chief executive, Bill Winters, is quoted on faster and more accurate detection of financial crime risk, and Standard Chartered's sanctions deployment dates to July 2018. Figures such as up to 90 percent better case resolution carry no named customer or method, and both banks are also investors.

Models were trained on millions of transactions from institutions worldwide and keep learning from investigator feedback, and anonymized datasets are pooled across institutions to counter bias. One bank's adjudications can therefore inform models other banks receive, with no opt out or boundary stated.

The privacy policy, updated July 2, 2026, reaches website visitors, job applicants and business contacts, not the screened customers and investigator decisions Iris processes for banks. Subprocessors are named only by category.

Security vulnerability disclosure has a footer page, and the privacy policy mentions technical and organizational measures without naming any. SOC 2 reports, ISO certifications and a trust portal do not appear.

A 2025 IMDA Spark recognition from Singapore's Infocomm Media Development Authority assesses technical strength and operational maturity, which is not a financial regulatory admission. Supervisors, statutes and instruments the product serves are not named.

Decisions are described as explainable, auditable and defensible, which matters when closing an alert on a named person. Name screening carries documented false positive differences across naming conventions and transliteration, and demographic testing or error analysis by population is unpublished.

Full traceability lets a bank asked by a supervisor why an alert closed produce the conclusion and its logic. Clients reporting no decision quality errors to date describes the past rather than a commitment, and the person screened has no route, largely because the regime forbids telling them.

Watchlist data is not supplied: clients bring their own sanctions and adverse media lists, and the models are presented as Silent Eight's own. Its language model provider for the Iris 7 agents, a subprocessor list and the hosting arrangement are unpublished.

Production use inside the screening operations of large banks shows the integration works, and case manager and risk data agents sit within Iris. Screening engines, case management platforms and core systems on either side are not named, and no public API documentation is linked.

Cloud, on premises and hybrid deployment are all stated in the FAQ, to meet jurisdictional and regulatory requirements. Hosting provider, region and tenancy model for the hosted service are unnamed.

Price stays unpublished, with a demo request as the way in. Alert volume, agent count and institution size are all candidate bases for charging, and none is confirmed.

Tier 1 banks are the stated audience, with Emirates NBD, Mashreq and First Abu Dhabi Bank among later partners, real reach through a small number of very large institutions. Insurers, smaller banks and fintechs are not addressed on the homepage or case studies page.