When a bank talks about AI on an earnings call, it usually means the virtual assistant in the app. Its patent applications are where the less-rehearsed ambitions show up — and in the week ending 20 April 2026, Bank of America (BAC) had a cluster of applications publish that put generative AI well past the chatbot, inside financial planning, corporate-structure analysis, and data protection. A published application is an application, not an enforceable right, and it reflects work filed roughly eighteen months earlier, so this is a dated but specific window onto where the bank has been pointing its AI engineering. The consistent signal across the set is that generative AI is being aimed at the substance of banking work, not just the front-door interface.

The most customer-facing application is US20260105522A1, which describes a generative-AI subsystem that builds a user-specific financial plan with tailored "resource checkpoints," then monitors the user's behavior against the plan and sends preemptive notifications when actions drift from it. The abstract describes the loop directly:

A behavior monitoring subsystem continuously tracks user actions and behavior, comparing them to the financial plan. Based on detected patterns, a notification subsystem can transmit preemptive notifications to the user, identifying potential issues or misalignments between the user's actions and their resource checkpoints, providing timely guidance to keep the user on track.— System and method for intelligent generation and tracking of resource checkpoints, US20260105522A1

This describes generative AI doing financial-planning work — generating a plan and nudging behavior toward it — rather than answering questions. Whether or not it ships in this form, an application of this shape points toward AI moving into advisory-adjacent territory that has historically belonged to human relationship managers and rule-based tools.

Generative AI for the institution, not just the customer

Two further applications aim generative AI at the bank's own decisions. US20260105215A1 describes generating real-time simulations to pre-evaluate "entity structure transformations" — analyzing target entities' capabilities and incapabilities and testing whether a proposed restructuring meets defined conditions, using generative AI. That is AI applied to corporate and organizational analysis, a back-office strategic function rather than a consumer feature. US20260105191A1 describes a generative-AI subsystem that learns the patterns in real customer data and produces synthetic "composite data," with a variable-obfuscation subsystem setting a data-transparency parameter based on how much a receiving entity is trusted. Synthetic data that stands in for protected records is a recurring need for any institution that wants to train and share AI systems without exposing the underlying customer information — and seeing a bank file on generating it indicates the data-protection side of AI adoption is itself a development priority.

The remaining applications point the same toolset inward, at running the bank's technology estate. US20260104913A1 describes an AI engine that monitors log data, infers each component's priority, predicts future log-usage requirements, and adjusts storage and retention accordingly — AI applied to the unglamorous but costly problem of log storage. US20260104678A1 describes robotic process automation paired with a machine-learning engine that derives network-event triggers and assigns bots to execute transmission protocols, and US20260106788A1 describes using machine learning to suppress redundant network-event data by grouping fractional events into aggregate events. These three describe AI and automation directed at the bank's network and operations layer — the infrastructure cost base that sits underneath everything customer-facing. Notably, several of the operations applications share inventors, which suggests one coordinated internal-operations program rather than unrelated filings.

Reading the signal, and its limits

Taken together, the week's applications sketch a two-front pattern that is coherent as fact: generative AI pushed toward customer-facing financial work and institutional analysis, and machine learning pushed toward the cost and reliability of the bank's own systems. That AI is now appearing in patent filings at the planning, restructuring, and data-protection layers — not only the assistant layer — is itself the signal worth marking, because patents are a leading and concrete indicator of where engineering attention has gone, distinct from the framing a bank chooses for investors.

There is a logic to why a bank, specifically, would file across both fronts at once. The customer-facing applications — US20260105522A1 on financial planning and US20260105215A1 on entity-structure simulation — are about applying AI to judgment-heavy work that has historically required people, which is where a large institution would look for productivity. The data-protection and operations applications — US20260105191A1 on composite synthetic data and US20260104913A1 on AI-based log reduction — are about the preconditions for doing any of that safely and affordably inside a regulated institution, where customer data cannot move freely and infrastructure cost is constantly scrutinized. A bank that wants to deploy generative AI at the front end has to solve the data-governance and cost problems at the back end first; filing on both in the same window is consistent with treating them as one program rather than two.

The interpretive limits matter, and they are specific to this format. A published application is not a granted patent; it confers no enforceable right, and some of these claims may narrow or never issue. Publication reflects the roughly eighteen-month statutory delay, so this is a view of prior R&D direction, not a current product disclosure — the records do not state which, if any, of these systems Bank of America has deployed, to how many customers, or with what result. Generative AI for planning, simulation, and synthetic data is also being filed across the financial sector, so a cluster here marks participation in a broad movement rather than a singular position. What the week establishes as fact is bounded and clear: in a single week of publications, the bank's set centered on generative AI inside financial planning, entity-structure simulation, and customer-data protection, alongside ML for its own network operations — a consistent forward-looking signal that AI development at a large bank has moved into the core functions, not just the interface.