Broadridge

The Clearwater precedent applied to the parent group. Strip every model and investor communications, proxy and governance processing, post trade infrastructure underpinning more than ten trillion dollars of daily trading, the wealth platform and the distributed ledger repo business all continue to run, because they did for decades. The learned line is real, in production and unusually well evidenced, which carries the build comfortably.

One qualification is recorded because it may warrant separate treatment later: the group's corporate bond venue is itself described as an artificial intelligence powered trading platform built around patented artificial intelligence, and graded on its own that business would likely reach B on this axis. It is treated here as a subsidiary line inside the parent rather than as a separate vendor.

The most complete oversight disclosure located in the trading and post trade lane, and the first vendor here to describe an agent that genuinely acts and then publish the architecture that bounds it. The bond trading agents can generate alerts, create trade tickets, select dealers, launch requests for quote and accept prices to execute automatically.

Against that the company publishes four named guardrails with their positions in the execution path stated: human in the loop approvals, policy driven limits on trade size and scope, explainability before every action, and full auditability of every action.

Explainability placed before the action rather than after it is the part that distinguishes this from every other vendor graded in the lane, because it makes the constraint operative rather than forensic, and limits on size and scope are a hard boundary on what the agent may do rather than a workflow description. The framing throughout is delegating selected tasks while keeping the trader in control.

Two things are recorded as unresolved rather than smoothed over: no default configuration or minimum limit is published, and because parameters are trader defined it is not fully clear whether human approval applies to every path or is one guardrail among several that a firm can configure.

Nothing about the models themselves. No accuracy or error rate, no validation methodology, no backtesting, no drift monitoring, no versioning and no external assessment, across a generative application answering bond questions and agents that can execute trades. Explainability and auditability are published and are credited on the oversight axis, where they function as controls on action rather than as evidence about model behaviour, and they are deliberately not counted twice here.

The distinction matters and is worth stating: explaining what an agent is about to do is not the same as publishing how well the underlying model performs, and this vendor does the first thoroughly and the second not at all.

The strongest deployment evidence in the platform tier and still no result attached to a named institution. What is genuinely better than its peers: the agentic capability is stated to be live in production rather than announced, and to have been refined through live deployments inside the company's own outsourcing operation across more than forty clients since 2024, which is production experience at a scale no other vendor on this roster claims.

Three global dealers are named as participants on the bond venue, and the repo platform's throughput is published with a period attached. Held at B on the same line as the rest of the tier: the one efficiency figure is a ceiling rather than a measurement and attaches to no institution, the named dealers are venue participants rather than referenced outcomes, and the throughput figures are the company's own volumes rather than a customer's result.

The vendor raises the pooling question itself and then does not answer it, which is the third clear instance of that shape in this sweep and the most explicit. It states that the agentic capabilities were refined through live deployments inside its own outsourcing operation across more than forty clients since 2024, so improvement through exposure to client operations is advertised as the reason the product is production ready.

Nothing states whose data produced that refinement, whether one client's exceptions and resolutions inform the agents another client receives, what the shared data ontology holds across clients, or what is excluded. The bond application is separately described as drawing on curated data and analytical models without saying whose. When a vendor sells that its agents got good by working on other people's operations, the question of whose operations is the whole question.

Nothing at product level in the material reviewed. No retention schedule, no deletion terms, no subprocessor list and no tenant separation statement, for a group holding shareholder records, trading activity and wealth account data.

The one architectural fact that bears on it, the choice between the vendor operating the process and the platform running inside the client's own infrastructure, determines custody of the data and is presented as a commercial option rather than as a data handling position.

No certification, attestation report, audit period or trust portal was located in the material reviewed, so the grade follows the standing rule that a credential must be found and named rather than assumed from a vendor's systemic importance.

This is unproven absence rather than evidenced absence and the queued check is high value: a listed processor underpinning trillions of dollars of daily settlement activity is examined and attested extensively, and the reason nothing surfaced is that such material sits in client due diligence channels and regulatory filings rather than on product pages. Recorded because the pattern now holds across the whole platform tier: the larger the processor, the less findable its security evidence.

Genuine regulatory standing held by the firm, disclosed in its own material, and only the second vendor on this roster to reach this grade. The corporate bond trading venue operates through a named group subsidiary acting as its broker dealer, which is a registered and supervised status carrying examination, capital requirements, conduct obligations and the possibility of sanction against the firm itself rather than only against its clients.

That places it inside the perimeter on the TrueML precedent, where a directly supervised and sanctionable firm is distinguished from every software vendor around it, and unlike the examination status recorded elsewhere in this sweep it is public and independently verifiable. Scope is stated precisely rather than overclaimed: the licence sits with the subsidiary operating the venue rather than attaching to every part of a group whose other businesses are technology and outsourcing. Worth noting for the wider index that a firm operating both the venue and the agents that trade on it holds a regulatory position that pure software suppliers of trading agents do not.

No published governance position, no fairness testing, no named framework and no statement on any artificial intelligence regulation. Fairness exposure is genuinely lower than for peers shipping consumer credit or collections capability, because the agents operate between institutional counterparties rather than deciding anything about a retail customer, and that is recorded as mitigation rather than as disclosure.

Two adjacent questions the material does not reach are worth flagging for later: whether agents acting across a venue could produce correlated behaviour among participants, and what governance applies to agents that a customer configures rather than the vendor building them.

The guardrails are published and the liability allocation is not, and that gap is sharper here than anywhere else in this sweep because of what the agents can do. An agent that accepts a price and executes automatically can produce a bad fill, a mispriced trade or an unintended position, and nothing published states who bears that loss, how an erroneous execution is unwound, or how responsibility divides between the venue, the agent platform, the firm that set the parameters and the trader who delegated the task.

The published limits on size and scope are a control on exposure rather than an answer to who pays when a permitted action turns out wrong. Recorded plainly: strong controls and silent liability are not the same thing.

No model, provider or version is named for the generative bond application or for the agents in production. Two dependencies at adjacent layers are named and neither is a model: a strategic investment in a named firm whose capability is being embedded into post trade operations, which is the naming the partner rung of the standing ladder, and an infrastructure and resiliency agreement with a named technology provider, which is the naming the cloud rung.

Both are more informative than silence and neither identifies what generates the output. The omission is more consequential than usual because the agents transact rather than advise, so a buyer assessing model risk on a trade executing agent cannot establish what is generating the decision to execute.

This vendor is not integrated with the market infrastructure, in several places it is the market infrastructure, which is the strongest form of the position that carried the other core suppliers on this roster. Post trade platforms underpin daily trading measured in trillions, the repo platform processed 7.2 trillion dollars in a single month, and the outsourcing arm runs client operations end to end rather than supplying software for the client to run.

The artificial intelligence work rests on what the company calls a completed financial services data ontology, which is a substrate artifact of the same class as publishing an integration schema or building on a named industry reference architecture, and it is what makes agents able to act across previously separate operational domains. A named infrastructure partner is disclosed for resiliency and platform capability.

Deployment choice is answered clearly and residency is not addressed. The two named paths are materially different rather than nominal variants: either the vendor operates the process on the client's behalf or the agentic platform is installed into the client's own infrastructure, and that distinction determines who holds the data.

Nothing published states region availability, residency commitments or how data is kept inside a jurisdiction for a group operating across twenty one countries in markets with divergent rules on where trading and investor records may sit. Graded C on the same basis as the other roster vendors that publish deployment choice without residency.

No pricing published. Two distinct commercial models are named, a full managed service in which the company runs operations end to end and a standalone deployment of the platform into the firm's own infrastructure, and naming the structure without naming a price is more than most of this roster offers and still not what this axis measures. A cost reduction ceiling is published in place of a cost, which tells a buyer what it might save and nothing about what it would pay.

Reach at the top of this index. The group serves sell side dealers, buy side institutions, wealth managers, corporate issuers and, through proxy and investor communications, retail shareholders at population scale, across twenty one countries, with platforms underpinning more than ten trillion dollars of daily securities trading and over seven billion communications a year.

The bond venue alone reports more than forty liquidity providers and over one hundred buy side institutions, with three global dealers named among participants. Buyer types therefore span trading, post trade operations, wealth, issuer services and outsourced operations, which are separate procurement centres with separate budgets rather than one market described broadly.