The AI FinTech Index Brief

The first two weeks in one line

The clearest pattern in the first two weeks of the change log is agents drafting documents that carry regulatory weight: suspicious activity report narratives, insurance claims files, investment proposals. The drafting moved to software. The signature, and the liability, did not.

This is the first issue of the Brief. It covers the change log’s opening period, August 3 to August 15: seven entries across seven vendors, every one Verified against a primary source. The log records shipped product, integration, regulatory, and evidence changes. Funding rounds, valuations, and awards are not logged, here or anywhere on this index.

The agents are drafting the filings

Unit21 introduced SAR Agents that draft FinCEN ready suspicious activity report narratives directly into a filing, synthesizing case data, alerts, and notes. The same release expanded real time payments screening to PEP and adverse media watchlists, aligned goAML filing workflows with the SAR generator, and updated alert auto assignment.

Federato launched Federato Claims, a claims module running from first notice of loss through payment, with an AI agent at intake capturing structured facts, checking applicable policy language, and flagging missing information. And Feathery introduced proposal generation for wealth management firms, with an assistant that drafts personalized investment proposals from client data, planning notes, and firm approved templates across Salesforce, eMoney, Holistiplan, and Morningstar.

A suspicious activity report, a claims file, and an investment proposal are not internal documents. Each one lands in front of a regulator, a policyholder, or a client, and each carries an obligation that belongs to a person, not a product. Three categories produced the same design in one fortnight: the draft moved to software, and the review and the signature stayed human.

Our read

The question that matters for a drafting agent is not whether the draft reads well. It is whether the reviewer can trace every statement in the draft back to a source before signing. An agent that cites its inputs shortens review. One that does not converts review into redoing the work, minus the attention.

Buyer question

For any agent drafting a regulated or client facing document, ask what the mandatory review step is, whether each statement traces to a source the reviewer can open, and whose name goes on the filing.

Market notes

The integrations landed on infrastructure buyers already run. Akur8 launched its Deploy Rating Accelerator on the Guidewire Marketplace, a mapping and transformation layer connecting PolicyCenter to its external rating engine for quote time rating calls and offline batch re rating. Theta Lake integrated with IPC One TradeDesk, which it says brings trade floor voice and digital communications under a single surveillance framework. And Built Technologies shipped its Summer 2026 release across construction loan administration: an AI Action Center for task generation and document extraction, a pay application drawer that mirrors the G702 form with automated retainage calculations, and stakeholder collaboration controls with four access levels.

The period’s one evidence entry: ThetaRay published results from Shift4’s deployment of its transaction monitoring platform, an 86 percent reduction in false positives and 70 percent more productive alerts while scaling past 200,000 monitored businesses. The figures come from the vendor’s own customer story, which is exactly the provenance to state when quoting them.

The inaugural note: shipped is not published

Building this log produced a finding worth more than any single entry. In the roster’s first full research pass, roughly two in five vendors publish no public changelog and no public API documentation, and several of the most active keep their release notes behind a customer login, including at least one that ships them weekly. The sector ships more than it publishes, so any change log for financial services AI is a record of the visible fraction, and how visible a vendor chooses to be is itself information.

That visibility gap matters because software can change materially between procurement reviews, security assessments, and contract renewals. If the record of those changes is unavailable, diligence quietly becomes archaeology.

The action item: one email

Ask your vendor where their changelog lives, and if the answer is behind a login, get release note access written into the contract. A public technical surface is a diligence signal in its own right. The vendors who document their changes in the open are also the ones whose claims this index verifies fastest, and that is not a coincidence.