Jack Henry

Core processing is the business, and a broad AI program is layered across it. Management reported 22 AI enabled products in market with more than 20 further capabilities identified for release within six months. The phrasing is the point: these are capabilities added to products rather than products that exist because of them. Take the models away and the company remains substantially intact.

The credit union and community bank cores, the digital banking platform serving more than 15.8 million registered users, the payments modules and the treasury estate are all deterministic systems of record that predate the models by decades. The financial crimes platform is the most model dependent product, and even there the analytics engine is a partner's rather than the company's own.

The security work with a cloud provider applies agentic defense to protect the estate rather than to sell intelligence. The annual report places AI within a technology strategy that also covers cloud rollout, serving both internal ways of working and product offerings.

The products are consistently assistive, and the company states that as a commitment. The digital banking assistant supplies suggested responses for staff and hands off between employee and automation rather than replacing the conversation. The fraud platform detects and alerts, and investigators make the disposition. Management commits publicly that clients always remain in control.

Real time interdiction exists, since the platform can stop a transaction before it leaves the institution, but under rules the institution sets. The item to examine is assisted drafting of suspicious activity report narratives. That document is what a regulator and potentially a court reads, written about a person who will never see it and cannot contest it. The stated benefit is giving the investigator back an hour to investigate further rather than removing the investigator. Nothing published states what the investigator reviews or attests before a drafted narrative becomes a filing.

Outcome disclosure is strong on one side and silent on the side that would make it interpretable. The published figure is well constructed: alert rates below one percent, representing roughly an 80 percent reduction in alerts reaching institutions, stated with the mechanism and the population rather than as a bare percentage. Narrative drafting saves at least an hour per investigation. What is missing is the counterpart.

An 80 percent reduction in alerts comes with no figure for how much real fraud detection was retained, and a reduction of that size is trivially achievable by detecting less. The reference to meaningful alerts gestures at retention without quantifying it. Validation methodology, false negative rate, and sample and observation period are also unpublished. There is no drift or retraining disclosure, and no model documentation an institution could put in front of its own examiners, which the engine belonging to a partner complicates further.

Evidence reported quarterly under securities law, which makes adoption claims auditable rather than assertable, and outcome figures that carry a mechanism. Adoption for the artificial intelligence products is stated with specificity and movement: 189 completed financial crimes platform installations as at 30 June 2026 with 57 more in progress, 183 new contracts across that platform and the faster payment modules over the year with 61 in the fourth quarter, 191 faster payment modules installed with 231 in progress, and digital platform signings of 219 for the year, up 24 percent, reaching more than 15.8 million registered users, up 11 percent.

Outcome evidence is quantified and explained rather than asserted: alert rates below one percent representing roughly an 80 percent reduction in alerts reaching institutions, and assisted drafting of regulatory report narratives saving at least an hour per investigation. Customers are named with named executives, including a federal savings bank and a community bank president. Independent recognition includes an analyst assessment naming the small business digital platform the leading one in its category and an industry award for the fraud platform.

A governance commitment made where it carries consequences, and two design choices that support it. Management stated in a quarterly earnings call that across all 22 AI enabled products and the further capabilities in development, strict risk management, compliance and governance frameworks would be maintained so that clients always remain in control. A statement to investors on a call carries securities exposure that a marketing page does not. The design choices reinforce it.

Adopting the Federal Reserve's published fraud classification model natively means classifications follow a public, regulator authored taxonomy rather than a proprietary scheme. That makes outcomes comparable across institutions and harder to quietly redefine. And the cloud collaboration applies agentic capability to defense, using AI to protect the estate rather than only to expose new surface. Nothing is published on model cards, evaluation methodology, red team results or incidents, and nothing states what accountholder data trains or contextualizes the models.

Regulatory compliance is sold as a product, and little is published about the company's own posture. The financial crimes platform is built around obligations its customers carry: screening against sanctions and watchlists, filing suspicious activity and currency transaction reports, and validating Bank Secrecy Act and anti money laundering compliance. The company demonstrably understands the regime its clients operate under.

That is compliance delivered to customers rather than privacy practiced by the vendor, and the two are not the same thing. No privacy program description, data processing disclosure, retention schedule or subprocessor list is public. GLBA appears nowhere on the company's public surfaces, though it processes deposit accounts for roughly 7,400 institutions governed by it.

The gap goes further because the fraud analytics run on a named third party engine and the security platform on a named cloud provider. Accountholder transaction data reaches parties beyond the vendor with no published description of what moves or under what terms.

A disclosure gap made conspicuous by what the company sells. No trust center, named certification, attestation report, penetration test summary or subprocessor list is public. The controls exist beyond reasonable doubt: a processor running deposit systems for roughly 7,400 supervised institutions is examined continually by those institutions and their regulators, and the company is a listed registrant with internal control obligations. The positioning is what makes the silence stand out.

The company is currently building a proprietary AI security platform to sell to its clients, designed around the strict compliance, regulatory and security requirements of community institutions. It publishes nothing verifiable about its own security credentials. A vendor asking institutions to trust it with their cyber defense has extra reason to publish what it holds. The practical effect falls hardest on the smallest clients, who lack the procurement leverage to demand documents a larger bank would simply require.

Jack Henry holds no license of its own, and its regulatory engagement is concrete. The financial crimes platform is built directly to named United States obligations rather than described as helping with compliance generally. It covers Bank Secrecy Act validation, suspicious activity and currency transaction report filing, and sanctions and watchlist screening. More distinctive is the native adoption of the Federal Reserve's published fraud classification model.

The vendor has taken a regulator authored taxonomy as its own classification scheme rather than inventing one, which makes an institution's reporting legible to its examiners without translation. The security platform is positioned against the compliance, regulatory and security requirements of community institutions, a segment supervised by multiple federal agencies.

As a listed registrant the company carries quarterly and annual reporting obligations and makes governance commitments publicly. Nothing is published about its own examinations as a bank service provider, and it takes no position under the EU AI Act, though its footprint is essentially domestic.

These determinations reach individuals, and nothing published addresses whether they land evenly. The fraud platform monitors accountholder transactions, flags suspicious activity and screens against watchlists. It now drafts the narrative that becomes a regulatory filing about a named person who is never told and cannot respond. Financial crime alerting has well documented uneven effects across geography, name origin and transliteration.

The added exposure is structural: the analytics engine belongs to a third party, so a community bank sits two removes from the model that judges its members and has less ability to interrogate it. Adopting the regulator's published classification taxonomy is a partial mitigation, because it standardizes what a fraud type means across institutions and makes patterns comparable. A shared vocabulary for classifying outcomes says nothing about whether the outcomes are distributed fairly. No fairness testing, differential performance analysis, model card or bias statement is published for any of the 22 AI enabled products.

No commercial instrument is published. There are no public terms of service or master agreement, no warranty or indemnity, and no liability cap or service level. Nothing states what an institution is owed when an AI product is wrong. Three exposures sit uncovered and they are not equivalent. The platform can stop a transaction before it leaves the institution, so a wrongly held payment is a direct customer harm.

Assisted drafting produces the narrative of a regulatory filing, so a defective draft that survives review becomes the institution's problem with its examiner rather than the vendor's. And because the analytics engine belongs to a named third party, an institution seeking recourse for a detection failure faces a chain rather than a counterparty. Nothing published describes how responsibility is divided between the platform and the engine behind it. Management's commitment that clients always remain in control is a governance posture, and it arguably places responsibility with the institution rather than allocating it contractually.

Jack Henry names the suppliers that actually matter, and that is the strongest part of its disclosure. The financial crimes platform is stated to be built through a partnership with a named specialist fraud analytics company. Both parties publish joint material about the collaboration and its results, so an institution knows precisely whose models judge its accountholders.

The security platform is built on a named cloud provider's agentic defense products under a relationship dating to 2022, and operational capabilities run on that provider's enterprise agent platform. Naming both the analytics engine and the cloud AI provider lets an institution assess those parties directly rather than inheriting undisclosed dependencies. Disclosure also makes one consequence visible.

The named analytics partner is itself an independent vendor in this market, so an institution running both that vendor directly and this platform may be paying twice for one model estate. It may also carry more concentrated exposure to a single engine than it realizes. No model family or version is named.

This vendor is the core for roughly 7,400 institutions, and it has made a structural choice that works against its own gatekeeping position. The integration network launched in July 2025 gives third party applications direct access to technical resources and open interfaces, explicitly removing the financial institution from the middle of the integration process, which is the opposite of the usual incumbent behaviour of routing every connection through the core provider and charging for the privilege.

The digital toolkit provides a public interface surface for the banking platform, and the company reports an open ecosystem of more than a thousand third party providers. The financial crimes platform connects across the company's own digital and payments estate and outward to third party systems including a major person to person network and custom institution systems. Payments reach spans the instant schemes with adoption reported by scheme.

For a community institution with no engineering capacity, being able to reach a fintech without the core provider standing in the way is the single most consequential integration property available, and this company publishes it as policy.

Cloud native and fully hosted, with the placement questions unaddressed. The financial crimes platform is described as fully hosted and built on public cloud, which the company presents as enabling real time data, more accurate reporting, open integrations and speed. The wider technology strategy is stated to include a rollout of cloud solutions. A named cloud provider relationship dating to 2022 underpins the security platform and operational AI work.

Nothing published closes the question for a regulated buyer. There is no region list, residency commitment or tenancy description, and no statement of what remains on institution controlled infrastructure versus what moves to the vendor or onward to its partners. That last point carries weight for this buyer set.

Community banks and credit unions with average assets around $1.2 to $1.3 billion have limited leverage in vendor negotiations and limited internal capacity to interrogate a hosting arrangement. Published clarity would matter more to them than to a large institution able to demand it.

No price, unit or tier appears for the core, the digital platform, the financial crimes product or any of the AI capabilities, on the company's site, in its investor materials or in the trade press. The securities filings supply a usable picture of the commercial model instead. The company reports contract signings and installations separately for the financial crimes platform and the faster payment modules, distinct from core client counts.

That shows these are modular products contracted individually rather than bundled into a core agreement, so adopting one does not require adopting everything and each carries its own negotiation. Segment revenue, client counts and growth rates are published quarterly with management commentary, so the size and health of the business unit doing the selling is visible. None of that is a price, and nothing indicates whether the AI capabilities carry incremental charges or arrive within existing licenses.

Coverage of the small institution tier is deep, and the company defines that tier precisely rather than gesturing at it. Roughly 7,400 community banks and credit unions are served. The annual report states the average core bank client at $1.29 billion in assets and the average core credit union client at 1.20 billion, which shows exactly which end of the market the company is built for. Credit unions are no side line: the credit union core is the dominant platform in its category.

Coverage runs across the estate those institutions actually need, from core processing and digital banking for retail and small business to payments, including the instant schemes with adoption reported by scheme. It also spans lending, treasury management, financial crime and Bank Secrecy Act compliance. What is absent is scale in the other direction. Nothing addresses large national or global institutions, and the geographic footprint is essentially domestic.