How Large Banks Use Generative AI: Bank by Bank (2026)

How are large banks using generative AI?

Large US banks mostly use generative AI through proprietary internal platforms that wrap outside models behind the bank's own security and data controls, then hand them to staff before customers. JPMorgan Chase's LLM Suite is available to about 250,000 employees and runs models from OpenAI and Anthropic; Citi's Stylus Workspaces has reached more than 180,000 people; Bank of America reports about 200,000 active users of its enterprise AI tools; Goldman Sachs put its GS AI Assistant in front of its whole workforce in 2025; BNY says every employee has access to its Eliza platform. The customer-facing uses are narrower and more recent: Morgan Stanley's adviser assistants, Bank of America's EricaAssist for call-centre staff, Truist's AI call summaries, Schwab's new Charley assistant and Capital One's Chat Concierge for car dealers.

Bank by bank

What proprietary generative AI platforms have large banks built?

Each row is what the bank itself or a named outlet has disclosed, with the date. Scale figures are quoted as reported and are not comparable across banks (access, active users and adoption are different measures). Platform names link through to the bank's page, where the full record is sourced.

Bank	Named platforms	What they do	Scale disclosed	Models
JPMorgan	LLM Suite	An in-house portal that gives employees large language models for drafting, summarising and analysis, updated about every eight weeks	About 250,000 employees have access (everyone except branch and call-centre staff), half of them using it roughly daily (Sep 2025)	OpenAI and Anthropic[1]
BofA	Enterprise AI tools; EricaAssist; Ask GPS	General-purpose and role-specific AI tools for staff; EricaAssist, with generative AI added in July 2026, guides call-centre employees in under three seconds; Ask GPS, built in-house on 3,200+ internal documents, answers payments staff	About 200,000 active users and more than 400,000 prompts a day (Q2 2026); EricaAssist used by 18,000+ employees and cut average call time by nearly a minute	Not disclosed[2][3][4]
Citi	Citi Stylus Workspaces; Citi Assist; Arc	Stylus Workspaces (agentic since September 2025) analyses complex data, researches and automates multi-step workflows; Arc is a platform for building and scaling AI agents	Stylus rolled out to more than 180,000 people; 65 million+ interactions with Citi's AI tools since December 2024, adoption above 87% (Sep 2026)	Not tied to a named model[5][6][7]
Goldman	GS AI Platform; GS AI Assistant	A firm-hosted platform that puts outside models behind Goldman's security layer; the Assistant is the conversational app on top of it	Assistant to about 10,000 staff in January 2025, then the whole workforce; 46,000 staff writing more than a million prompts a month (Jul 2025); 12,000 developers on GitHub Copilot	OpenAI and Google, per Fortune[8][9][10]
Morgan Stanley	AI @ Morgan Stanley Assistant; Debrief; AskResearchGPT	The Assistant answers advisers from the firm's intellectual capital; Debrief writes meeting notes, action items and follow-up emails with client consent; AskResearchGPT searches Morgan Stanley Research for bankers, traders and analysts	Assistant used by 98% of financial-adviser teams (Jun 2024); AskResearchGPT covers 70,000+ research reports a year	OpenAI[11][12]
Wells Fargo	Fargo; Google Agentspace; AI Teammate	Fargo is the customer assistant in the mobile app; in August 2025 Wells began rolling Google's agent tools out to staff; AI Teammate (July 2026) adds plain-language chat to advisers' Advisor Gateway	Fargo past 1 billion interactions (Mar 2026); Agentspace rollout aimed at 215,000 employees	Google (Agentspace)[13][14][15]
BNY	Eliza	An enterprise AI platform with a menu of models on which any employee can build solutions and agents	All employees have access; about 220 enterprise AI solutions in production and about 140 “digital employees” (Q1 2026)	OpenAI and Google, among others[16][17]
Capital One	Chat Concierge	Capital One's first proprietary multi-agent assistant: on car dealers' websites it answers questions, compares vehicles and books test drives	Launched March 2025 for dealers on Capital One's auto platform	Meta's open-source Llama, customised with Capital One data[18]
Truist	AI Call Summaries; Truist Assist	Generative AI writes the summary of each care-centre call; Truist Assist is the digital assistant for clients	More than 4.5 million summaries in Q2 2026, covering over 70% of inbound care-centre interactions and saving more than 36,000 hours	Not disclosed[19]
Schwab	Schwab Knowledge Assistant; Portfolio Insights; Charley	Knowledge Assistant gives client-service reps cited answers; Portfolio Insights summarises account performance for clients; Charley is a generative-AI assistant for retail clients	Portfolio Insights to all self-directed US clients from May 2026; Charley announced September 30, 2026, phased in from October	Not disclosed[20][21][22]
U.S. Bank	Design Assistant; developer-portal assistant	Internal tools from U.S. Bank's AI centre of excellence: a design assistant for its designers and a generative-AI assistant for embedded-banking clients using its APIs	No usage figures disclosed	Not disclosed[23][24]
PNC	“AI factory” (in development)	PNC's plan to run its own GPUs and data centres and its own large and small language models instead of renting them	Announced June 2026; not yet a live product	Its own models (planned)[25]
Amex	Agentic Commerce Experiences (ACE) developer kit	Lets businesses build Amex payments into AI-agent shopping, with agent registration, tokenized payment credentials and cart context; Amex also plans Agent Purchase Protection	Released to select developers on April 14, 2026; features under development	Not disclosed[26]

The pattern

Do large banks build their own AI models?

Mostly not from scratch. The common design is a proprietary platform around models bought from OpenAI, Anthropic, Google or Meta: JPMorgan's LLM Suite uses OpenAI and Anthropic models, Goldman's GS AI Platform puts OpenAI and Google models behind its own security layer, BNY's Eliza offers a menu of models, and Capital One customises Meta's open-source Llama for Chat Concierge. PNC is the exception in the making: it has said it will run its own GPUs and its own large and small language models.

Employees come first. The largest deployments by number of users are all internal (LLM Suite, Stylus, Bank of America's enterprise tools, the GS AI Assistant, Eliza). Where generative AI faces customers it usually sits behind a person (Morgan Stanley's advisers, EricaAssist for call-centre staff, Truist's call summaries) or in a bounded setting (Chat Concierge on dealer websites). Schwab's Charley, announced on September 30, 2026, is one of the first general-purpose generative assistants offered directly to retail clients by a large US firm.

Common questions

What is JPMorgan's LLM Suite?

LLM Suite is JPMorgan Chase's in-house generative AI platform. CNBC reported in September 2025 that about 250,000 employees, the whole workforce except branch and call-centre staff, had access, that half used it roughly daily, and that it ran models from OpenAI and Anthropic, updated about every eight weeks.

Which bank has the most employees using generative AI?

On disclosed figures, JPMorgan Chase (about 250,000 with access to LLM Suite, September 2025), Bank of America (about 200,000 active users of its enterprise AI tools, Q2 2026) and Citigroup (Stylus Workspaces rolled out to more than 180,000 people, September 2026). The measures differ, so the order is approximate.

Do banks use ChatGPT?

Several use OpenAI's models inside their own platforms rather than the public ChatGPT app: JPMorgan's LLM Suite, Morgan Stanley's adviser assistants and BNY's Eliza all run OpenAI models among others. Goldman Sachs uses OpenAI and Google models behind its own security layer.

How is this different from the AI index?

The AI index scores the 120 largest US banks on their whole public AI record. This page lists only the generative AI platforms large banks have named, what they do and the scale they have disclosed.

Sources

Also on this site

How are large banks using generative AI?The proprietary platforms, bank by bank.