Verified
Visa
Piloting- Payments
- ~34,800 employees
- 89% of staff use AI (Apr 2026)
- 150+ AI apps
- Agentic Toolchain built in-house
- No AI return published
Visa deploys AI in its own risk models, which score every transaction on VisaNet, and in generative and agentic tools for staff, built on OpenAI, Anthropic and Google models. Visa said in Apr 2026 that 89% of its employees actively use AI, and in Oct 2026 that more than 60% of its product and engineering teams build with its in-house Agentic Toolchain. Visa announced about 2,600 job cuts in Jul 2026, saying AI is helping change how work gets done, but has published no return it credits to AI alone.
usa.visa.comTheir approach to AI
- The platform
- Visa built its own data and AI platform, which Rajat Taneja said cost $3.5 billion over 5 years (Oct 2025). Visa also runs Secure ChatGPT, a multi-model staff interface on Microsoft Azure, and the Visa Agentic Toolchain for building products (fiscal 2026 Form 10-K).
- The models
- Visa staff use Anthropic's Claude most, closely followed by OpenAI's ChatGPT, with Google's Gemini also popular (Apr 2026). Visa also trains its own risk models, including a Large Transaction Model built on billions of transactions.
- The reach
- Visa told Business Insider in Apr 2026 that 89% of its employees are active AI users and 44% are power users, averaging at least 25 prompts a day on 15 days a month.
- The catch
- Visa has published no business return it credits to AI alone. Visa's fiscal 2026 Form 10-K books the Jul 2026 job cuts as an efficiency program, its fraud figures count losses prevented for its clients, and its Agentic Toolchain gains are speed measures.
Snapshot
- Employees
- ~34,800
- Sep 30, 2026, up from about 34,100 in fiscal 2025, after about 2,600 job cuts announced in Jul 2026
- Technology investment
- ~$12B over 5 yearsest.
- Rajat Taneja, Oct 2025; Visa cited $15B in network security and resilience over 5 years in Oct 2026
- AI/ML headcount
- 2,500+ engineers on AI
- Oct 2025, before the Jul 2026 job cuts, which fell mostly on technology and product teams
- Who owns AI
- Rajat Taneja, President, Technology
- Executive officer over product engineering, global IT and operations infrastructure
- Platform
- Data and AI platform, built in-house
- Plus Secure ChatGPT on Microsoft Azure and the Visa Agentic Toolchain (fiscal 2026)
- Models
- Anthropic, OpenAI, Google, open-source
- Claude most used by staff (Apr 2026); own risk models, including a Large Transaction Model
- Reach
- 89% active AI users
- Apr 2026; 44% of employees are power users
- Use cases
- 150+ AI-powered applications
- Jul 2026; Visa cited more than 100 internal AI business applications in Oct 2025
Key findings
Visa has published no business return it credits to AI alone. Visa announced about 2,600 job cuts, 7% of staff, on Jul 28, 2026 with $563 million of severance, and Ryan McInerney wrote that AI is helping accelerate the change. Visa's fiscal 2026 Form 10-K books the severance as an efficiency program, and Visa's headcount still rose to about 34,800 by Sep 30, 2026.
Visa told Business Insider in Apr 2026 that 89% of its employees are active AI users and 44% are power users. Visa told Business Insider that its staff used 1.9 trillion AI tokens in Mar 2026, double the Feb 2026 figure, and Anthropic's Claude was their most popular assistant.
Visa is replacing product teams of 10 or more with agentic squads of 2 to 4. Ryan McInerney said in Jul 2026 that teams on the Visa Agentic Toolchain post 80% more code commits and 65%+ faster feature development. More than 60% of Visa's product and engineering teams worked this way by Oct 2026.
Visa's agentic commerce program has added Visa Intelligent Commerce (Apr 2025), the Trusted Agent Protocol with Cloudflare (Oct 2025), an OpenAI partnership (Jun 2026) and work on Anthropic's commerce blueprint (Sep 2026). Nearly 50 AI partners were live on Visa Intelligent Commerce by Oct 2026, but Ryan McInerney said agentic commerce will remain very small for Visa in fiscal 2027.
Visa tested Anthropic's Claude Mythos Preview on its payment network through Project Glasswing and open-sourced the Visa Vulnerability Agentic Harness on Jun 10, 2026. Visa agreed on Aug 3, 2026 to buy BioCatch, a provider of behavioral-first, multi-signal fraud intelligence, for $2.4 billion, after buying Featurespace for $946 million in Dec 2024.
Where Visa sits
- Stage
- Piloting
Visa has reported no business return it credits to AI alone, so its stage rests on rollout evidence and sits at Piloting. Visa announced about 2,600 job cuts on Jul 28, 2026, with $563 million of severance, and CEO Ryan McInerney wrote that AI is helping accelerate the change. Visa's fiscal 2026 Form 10-K credits the severance to actions to drive operational efficiencies and reinvest. Bloomberg reported that AI was not the only factor, and Visa's headcount rose to about 34,800 by Sep 30, 2026. Visa's fraud figures count losses prevented for its clients, and its Agentic Toolchain results (80% more code commits, 65%+ faster feature development, Jul 2026) are speed measures. On rollout, Visa told Business Insider in Apr 2026 that 89% of its employees are active AI users and 44% are power users. Rajat Taneja, the executive officer who is President, Technology, owns the platform. Visa falls short of Scaling because it reports no return it credits to AI in any function and no firm-level AI figure.
How the stages are defined - Ownership
- Rajat Taneja, President, Technology and an executive officer
- Platform
- Hybrid: Data and AI platform and Agentic Toolchain built in-house; assistants bought
- Models
- Anthropic, OpenAI and Google for staff, plus Visa's own risk models
- Hosting
- Secure ChatGPT runs on Microsoft Azure behind Visa's firewall; VisaNet risk scoring runs in Visa's own data centers; other AI clouds not named
- Reach
- 89% of employees active AI users (Apr 2026)
How they deploy
A decade of platform work before generative AI
Rajat Taneja told Fortune in Oct 2025 that Visa rewrote its data platform from scratch about a decade earlier to use deep neural nets, with layers for machine learning workbenches and faster model training. Taneja said Visa began working with generative AI around 2018 or 2019 and that every Visa employee had access to a generative AI model by Feb 2023. Taneja put the cost of building Visa's AI platform at $3.5 billion over 5 years, out of about $12 billion Visa spent on technology in the same period.
Many models, no lock-in
Rajat Taneja told Fortune in Feb 2025 that Visa works with OpenAI, Anthropic's Claude, Google and open-source models from IBM, Meta and Mistral AI, and matches the model to each job. Taneja said Visa wants to avoid vendor lock-in and favors models that are transparent, work on fairness and let Visa fine-tune them. Visa's open-source vulnerability harness follows the same pattern: an abstraction layer lets Visa swap or combine model providers without changing its controls (VentureBeat, Jul 2026).
Agentic squads in product development
Ryan McInerney said on Jul 28, 2026 that Visa is moving beyond AI assistance to AI that executes work under supervision, re-forming product development teams of 10 or more into agentic squads of 2 to 4. Visa's fiscal 2026 Form 10-K describes the Visa Agentic Toolchain as an AI pipeline run with human oversight. The Agentic Toolchain writes and critiques requirements, builds prototypes, generates and corrects code, supports releases and captures feedback. McInerney said on Oct 27, 2026 that more than 60% of Visa's product development and engineering teams had adopted the toolchain, with the intent of reaching 100% "this year".
Agentic commerce as the growth bet
Ryan McInerney said in Jul 2026 that Visa believes AI and agentic commerce will expand its addressable market. Visa's fiscal 2026 Form 10-K calls Visa Intelligent Commerce its agentic platform, built with frontier AI labs, sellers and issuers. Rubail Birwadker, Visa's global head of growth, told Axios in Jun 2026 that more than 1 in 5 transactions are influenced by what shoppers learn through large language models (SiliconANGLE). McInerney told analysts on Oct 27, 2026 that agentic commerce remains very small against the size of Visa's business for fiscal 2027.
AI inside the risk products, built and bought
Visa scores every VisaNet transaction in flight with Visa Advanced Authorization. Visa is adding its Large Transaction Model, trained on billions of transactions, to its risk services to improve fraud detection and cut false declines (fiscal 2026 Form 10-K). Visa bought Featurespace, a real-time AI payments protection company, for $946 million in Dec 2024. Visa agreed in Aug 2026 to buy BioCatch for $2.4 billion in cash and expects the deal to close by the end of its fiscal second quarter of 2027.
Rewarding results over usage
Visa rewards teams that show they used AI to get work done faster. Rajat Taneja told Business Insider in Apr 2026 that the measure is the volume of impact, not token counts. One award went to a team that used Claude Sonnet to ship a new API in under 6 days. Visa's Dec 2025 proxy statement lists the use of generative AI to raise internal productivity among the fiscal 2025 performance results for both Taneja and Paul Fabara.
- Deployment model
- Visa runs AI through a central technology organization with product and risk work in the businesses. Rajat Taneja, President, Technology, is responsible for technology strategy, product engineering, global IT and operations infrastructure, and for integrating engineering and product teams (Dec 2025 proxy statement). Andres Vives runs the Global Data Office as Chief Data Officer. Jack Forestell leads product and strategy, including Visa Intelligent Commerce, and Andrew Torre runs value-added services, where most of Visa's AI risk products are sold. Visa has not published where the data and AI platform team or the chief information security officer report. Visa has also not said who has led the platform team since its former head, Sam Hamilton, left Visa by Mar 2026. Visa staff reach AI through an internal generative AI hub and an AI assistant for employees (fiscal 2026 Form 10-K), and through Secure ChatGPT on Microsoft Azure. Visa gives GitHub Copilot to all developers (Mar 2025) and Microsoft Copilot licenses to most employees (Oct 2025), and staff also use Claude, ChatGPT and Gemini (Apr 2026).
- Model providers
- Visa uses models from several providers: Rajat Taneja named OpenAI, Anthropic, Google and open-source models from IBM, Meta and Mistral AI in Feb 2025. Taneja told Business Insider in Apr 2026 that Claude is the most popular assistant among Visa staff, often swapping places with ChatGPT, and that Google's Gemini is also widely used. Visa's Secure ChatGPT puts GPT, Mistral, Claude, Llama, Gemini and IBM Granite models behind one interface (Sam Hamilton, Mar 2025). Visa staff pick smaller open-source models such as Mistral for light fine-tuning, or OpenAI's reasoning models for harder work. Visa gives all developers GitHub Copilot, joined Microsoft's Copilot pilot early and has since bought Microsoft Copilot licenses for the vast majority of employees (Oct 2025). Visa tested Anthropic's Claude Mythos Preview through Project Glasswing in 2026, and its open-sourced vulnerability harness runs on Anthropic or OpenAI-compatible models. For risk, Visa trains its own models, including the deep learning detection models behind its fraud scoring and a Large Transaction Model described in its fiscal 2026 Form 10-K.
- Data and infrastructure
- Visa processes payments on VisaNet through its own global data centers, which have redundant connectivity, power and cooling (fiscal 2026 Form 10-K). Visa Advanced Authorization scores every VisaNet transaction in real time. Sam Hamilton said in Mar 2025 that Visa's Secure ChatGPT runs internally on Microsoft Azure behind a firewall, with data loss prevention screening so sensitive data does not leave Visa's systems. Hamilton said Visa's data platform holds hundreds of petabytes in a data lake, encrypted at rest and in transit. Visa's fiscal 2026 Form 10-K says its use of AI depends in part on third-party tools, models and infrastructure that Visa does not fully control. Visa has not said which clouds host its Agentic Toolchain or other generative AI tools. Visa said in Oct 2026 that it had invested $15 billion in network security and resilience over 5 years. Rajat Taneja said in Feb 2025 that Visa set rules after ChatGPT's release to protect company data, avoid copyright infringement and keep the models it uses compliant.
Platforms and internal tools
Visa data and AI platformRebuilt from about 2015
Visa's data and AI platform is the in-house stack, rebuilt over several years, on which Visa trains and governs its models. The platform includes machine learning workbenches, a model catalog and real-time drift alerts. Rajat Taneja said in Oct 2025 that Visa spent $3.5 billion over 5 years building the platform. Taneja showed Fortune an AI observatory dashboard that tracks each model's compliance, risk profile, privacy and security.
Secure ChatGPT and the generative AI hubModel access for all staff by Feb 2023
Secure ChatGPT is Visa's internal generative AI interface, run on Microsoft Azure behind Visa's firewall, and screens inputs and outputs with data loss prevention (Sam Hamilton, then Visa's senior vice president of data and AI, Mar 2025). Despite its name, Secure ChatGPT offers models from OpenAI, Mistral AI, Anthropic, Meta, Google and IBM, and serves as a retrieval and model service for internal apps. Visa's fiscal 2026 Form 10-K describes an internal generative AI hub and an AI assistant for employees. Visa has not said whether the hub is the same service.
- Users
- 89% of employees active AI users, 44% power users (Apr 2026)
- Models
- OpenAI GPT, Mistral, Anthropic Claude, Meta Llama, Google Gemini, IBM Granite
Copilot coding and productivity assistantsBy Feb 2025
Visa gives all of its developers GitHub Copilot for coding and unit testing, which Sam Hamilton said significantly raised code test coverage (Mar 2025). Visa joined Microsoft's early Copilot pilot and has since bought Microsoft Copilot licenses for the vast majority of employees (Oct 2025).
- Models
- GitHub Copilot, Microsoft Copilot
Visa Agentic ToolchainFiscal 2026
The Visa Agentic Toolchain is the AI pipeline, run with human oversight, that Visa's agentic squads of 2 to 4 people use to build products. The Agentic Toolchain develops and critiques requirements, creates prototypes, generates and corrects code, supports releases and captures post-launch feedback.
- Users
- More than 60% of product development and engineering teams (Oct 2026)
Visa Intelligent CommerceApr 30, 2025
Visa Intelligent Commerce opens the Visa network to developers building AI shopping agents. Its tools include tokenized agent credentials, spending controls, the Trusted Agent Protocol, an agent directory and Agent Score for sellers. Visa said in Oct 2026 that nearly 50 AI partners are live in production, including Shopify's Shop Pay, and that the number is expected to double in the coming months.
Visa Advanced Authorization and the Large Transaction Model
Visa Advanced Authorization gives issuers an in-flight risk score for every VisaNet transaction. Visa is adding its Large Transaction Model, trained on billions of transactions, to Visa Advanced Authorization and Visa Decision Manager to sharpen fraud detection, improve authorization and reduce false declines (fiscal 2026 Form 10-K).
Visa Vulnerability Agentic HarnessJun 10, 2026 (open source)
The Visa Vulnerability Agentic Harness is a governed pipeline, in its fifth generation, that directs frontier models through 4 phases and 11 stages with deterministic controls and human review. The harness covers threat modeling, verification, exploit chain synthesis, remediation and fix validation. Visa built it while testing Claude Mythos Preview and released it on GitHub under an open-source license.
- Models
- Anthropic Claude or OpenAI-compatible models; fixes require Anthropic models
Vendors and partners
AnthropicBy Feb 2025
Claude for staff, Claude Mythos Preview testing through Project Glasswing, and the commerce agent blueprint · Claude was the most popular assistant among Visa staff in Apr 2026, and Visa is one of the partners named on Anthropic's commerce agent blueprint (Sep 2026).
OpenAIBy Feb 2025
Internal ChatGPT for staff and an agentic commerce partnership · Visa and OpenAI agreed on Jun 10, 2026 to integrate Visa payments into OpenAI experiences and to explore enterprise uses, including developer tools powered by Codex.
Microsoft
Azure hosts Secure ChatGPT; Microsoft Copilot and GitHub Copilot for staff; Visa Intelligent Commerce collaborator · Secure ChatGPT runs on Microsoft Azure behind Visa's firewall (Mar 2025), and Visa has bought Microsoft Copilot licenses for the vast majority of employees (Oct 2025).
Google
Gemini models and assistant for staff · Rajat Taneja named Google among Visa's model providers in Feb 2025, and Gemini was widely used internally in Apr 2026.
CloudflareOct 2025
Co-developer of the Trusted Agent Protocol
Meta
Llama models in Secure ChatGPT; payments on Facebook and Instagram through Visa Intelligent Commerce · Ryan McInerney said in Jul 2026 that Visa is enabling new ways to pay on Facebook and Instagram powered by Visa Intelligent Commerce.
IBM and Mistral AIBy Feb 2025
Granite and Mistral models in Secure ChatGPT; Visa Intelligent Commerce collaborators
FeaturespaceDec 2024
Acquired real-time AI payments protection company · Visa paid $946 million and sells Featurespace as a platform for clients to combine risk signals.
BioCatchAug 2026
Provider of behavioral-first, multi-signal fraud intelligence that Visa agreed to buy · Visa agreed on Aug 3, 2026 to buy BioCatch for $2.4 billion in cash, with closing expected by the end of its fiscal second quarter of 2027.
Leadership
AI, data and platform leaders
Executives
AI org map
Central technology and data and AI platform; AI products built and sold by product and value-added services teams
Visa places AI under Rajat Taneja, President, Technology, one of the executive officers listed in its Dec 2025 proxy statement. Taneja's remit covers technology strategy, product engineering, global IT and operations infrastructure, and the integration of engineering and product teams. Visa's Chief Data Officer, Andres Vives, leads the Global Data Office. Jack Forestell runs product and strategy, and Andrew Torre runs value-added services. Visa's board oversees AI as part of strategy, and its Audit and Risk Committee oversees AI risks.
Visa's AI work sits mainly in its technology organization under Rajat Taneja. Fortune described that organization as a 12,000-strong team in Feb 2025, and reported more than 2,500 engineers working on AI in Oct 2025. Visa's product, risk, value-added services and consulting teams build and sell AI products on that platform, and Visa Consulting & Analytics runs an AI advisory practice for clients. Visa has not published an AI headcount since its Jul 2026 job cuts, which fell mostly on technology and product teams.
Reporting lines
Chief Executive Officer
Technology
Rajat Taneja, President, Technology
Executive officer; product engineering, global IT and operations infrastructure (Dec 2025 proxy statement); the line to the CEO is inferred, not published
Data and AI platform
Former head Sam Hamilton had left Visa by Mar 2026; Visa has not named a successor or published the reporting line
Cybersecurity
Subra Kumaraswamy, CISO
Reporting line inferred; Kumaraswamy co-wrote the Jun 2026 harness release with Rajat Taneja
Product and Strategy
Owns Visa Intelligent Commerce and the agentic commerce partnerships; the line to the CEO is inferred
Finance
Chris Suh, CFO
Global Data Office
Andres Vives, Chief Data Officer
Leads data and AI-related work including governance and AI policy; Visa has not published where the office reports
Visa's Dec 2025 proxy statement lists Rajat Taneja, Chris Suh and Paul Fabara among the executive officers besides Ryan McInerney. Jack Forestell, Andres Vives and Andrew Torre are not on that list, and Visa has not published their reporting lines.
Where the AI workforce sits
Technology organization and the data and AI platform
Estimate: Visa gives no breakdown. Fortune put more than 2,500 engineers on AI in Oct 2025 inside a technology team of about 12,000, and Visa reports its deepest AI use in product development and engineering.
~70-85%Product, risk, value-added services and consulting
Estimate: the remainder, including Visa Consulting & Analytics data scientists and the teams behind Visa's AI risk and dispute products.
~15-30%
Named people
Use cases
Ryan McInerney said Visa had more than 150 AI-powered applications in Jul 2026. Visa's AI runs in fraud and risk scoring on VisaNet, agentic commerce, dispute handling, consulting, client service, cybersecurity and Visa's own product development. Visa has published no business return it credits to AI for any named use case. Visa's fraud figures count losses prevented for issuers and sellers, and its Agentic Toolchain results are speed and throughput measures. Visa's value-added services revenue rose 30% in constant dollars to about $14 billion in fiscal 2026, but Visa does not split it by AI. At firm level, Visa announced about 2,600 job cuts in Jul 2026 with AI cited as one factor, and booked the $563 million of severance as an efficiency program.
| Use case | Area | Status | ROI |
|---|---|---|---|
| Visa Advanced Authorization Visa Advanced Authorization scores every transaction on VisaNet in flight and sends the score to the issuer, which decides whether to approve. Visa told Fortune in Feb 2025 that generative AI and deep learning detection models help it prevent about $40 billion in fraud attempts a year. | Risk and security | In production | |
| Large Transaction Model Visa's Large Transaction Model is trained on billions of transactions and is being added to Visa Advanced Authorization and Visa Decision Manager to improve fraud detection and authorization and reduce false declines. | Risk and security | In production | |
| Visa Protect and account attack scoring The Visa Protect suite uses AI, machine learning and network intelligence to help issuers and sellers fight fraud. The suite includes Visa Account Attack Intelligence, which uses generative AI to score enumeration attacks (automated testing of payment fields), and Visa Protect for account-to-account payments. | Risk and security | In production | |
| Featurespace and BioCatch fraud intelligence Visa sells Featurespace, bought in Dec 2024, as a platform for clients to combine risk signals. Visa agreed in Aug 2026 to buy BioCatch, whose behavioral models protect about 760 million users on 1.8 billion devices for more than 100 of the largest banks. | Risk and security | In production | |
| AI dispute resolution suite Visa launched 6 AI tools for disputes on Apr 1, 2026. The tools include a recovery manager that uses generative AI to automate responses and predict the likelihood of winning, and a document analyzer that summarizes merchant evidence in a structured format. | Value-added services | Announced | |
| Visa Intelligent Authorization Visa Intelligent Authorization, launched in fiscal 2026, uses analytics and AI to help acquirers and payment service providers raise authorization rates and cut payment friction. | Value-added services | In production | |
| Visa Intelligent Commerce and the Trusted Agent Protocol Visa Intelligent Commerce gives AI agents tokenized credentials and spending controls, and the Trusted Agent Protocol lets merchants verify agents with cryptographic signatures. Partners completed hundreds of agent-initiated transactions by Dec 2025, and nearly 50 AI partners were live in production by Oct 2026. | Agentic commerce | In production | |
| Payments inside OpenAI and on Anthropic's commerce blueprint Visa agreed with OpenAI on Jun 10, 2026 to let agents start Visa payments within user-set limits, with tokenized credentials and real-time fraud monitoring. Neither company gave a launch date for the OpenAI experience. In Sep 2026, Visa joined Anthropic's commerce agent blueprint as a partner. | Agentic commerce | Announced | |
| AI Financial Assistant Visa is developing an AI Financial Assistant that banks can white-label to give cardholders AI-powered financial insights from the bank's data and Visa's network data inside the bank's own app. | Issuing solutions | Announced | |
| AI-assisted consulting projects Visa Consulting & Analytics uses AI-enabled tools in its advisory work. Chris Suh said AI helped the unit deliver 1,200 consulting projects for more than 700 clients in the quarter to Jun 2026, more than in all of 2019. | Visa Consulting & Analytics | In production | |
| Visa Agentic Toolchain Visa's agentic squads of 2 to 4 people use the Visa Agentic Toolchain to write requirements, prototype, code and ship. Ryan McInerney said in Jul 2026 that the squads post 80% more code commits, cut the time to define requirements from 30 days to 5 and develop features 65%+ faster. McInerney said in Oct 2026 that 12 people built the platform behind Visa's Open USD stablecoin support in 7 weeks. | Product development and engineering | In production | |
| Generative AI client support Visa's client services team uses Secure ChatGPT to answer country-specific policy questions by retrieving Visa's source documents and citing them. Sam Hamilton said in Mar 2025 that such a search once took hours or days and now takes minutes. Ryan McInerney named client service among the first areas where Visa deployed AI. | Client service | In production | |
| Seller monitoring Visa uses AI-powered monitoring tools to detect sellers that conceal the true nature of their business to avoid its compliance rules, and requires acquirers to act when it suspects illegal activity (Dec 2025 proxy statement). | Compliance | In production | |
| Agentic vulnerability hunting Visa pointed Anthropic's Claude Mythos Preview at its payment network through Project Glasswing. The model chained minor weaknesses into working exploit paths, some rated critical, which Visa says its zero-trust controls stopped. Visa now runs the open-sourced Visa Vulnerability Agentic Harness and tracks a Mean Time to Adapt measure. | Cybersecurity | In production | |
| Secure ChatGPT, employee assistant and Copilot Visa employees use Secure ChatGPT, an internal generative AI hub and AI assistant, Microsoft Copilot, GitHub Copilot for developers, and Claude, ChatGPT and Gemini. In Apr 2026, software engineering was Visa's heaviest token user, and marketing was among the other functions taking AI up. | Firmwide | In production |
Adoption
Visa spread AI in layers: neural-network fraud scoring in VisaNet for years, a rebuilt data and AI platform, generative AI for every employee by Feb 2023, and agentic tooling for product teams in fiscal 2026. Nearly all Visa staff had a secure internal ChatGPT by Feb 2025, and Visa had bought Microsoft Copilot licenses for the vast majority of employees by Oct 2025. Visa told Business Insider in Apr 2026 that 89% of employees are active AI users and 44% are power users, and that monthly token use doubled from Feb to Mar 2026, to 1.9 trillion. Visa's application counts measure different things. PYMNTS reported more than 500 generative AI applications in 2024, Fortune more than 100 internal AI business applications in Oct 2025, and Ryan McInerney more than 150 AI-powered applications in Jul 2026. By Oct 2026, more than 60% of Visa's product development and engineering teams worked through the Agentic Toolchain. McInerney said adoption outside product and engineering is good but behind.
By the numbers
Rollout, first pilot to today
About 2015
Visa began rewriting its data platform from scratch for deep neural nets, with layers for machine learning workbenches and faster training, according to Rajat Taneja.
Feb 2023
Every Visa employee had access to a generative AI model, and Visa joined Microsoft's Copilot pilot before buying licenses for most staff.
2024
Visa reported more than 500 generative AI applications in use, including a code security bug finder and chatbots that act as subject matter experts.
Dec 2024
Visa bought Featurespace, a real-time AI payments protection company, for $946 million.
Feb 2025
Visa had given nearly all its workforce a secure internal version of ChatGPT and held Microsoft Copilot licenses for software developers and the sales team.
Apr 30, 2025
Visa launched Visa Intelligent Commerce with Anthropic, IBM, Microsoft, Mistral AI, OpenAI, Perplexity, Samsung and Stripe among its collaborators.
Oct 2025
Rajat Taneja said Visa had built more than 100 internal AI business applications and had more than 2,500 engineers working on AI.
Apr 2026
Visa told Business Insider that 89% of its employees are active AI users and 44% are power users, and that monthly token use reached 1.9 trillion in Mar 2026.
Jul 28, 2026
Ryan McInerney said Visa is re-forming product teams into agentic squads of 2 to 4 and has more than 150 AI-powered applications.
Oct 27, 2026
Ryan McInerney said more than 60% of Visa's product development and engineering teams had adopted the agentic way of working, with the intent of reaching 100% "this year".
Economics & controls
Spend and ROI
Technology budget
Visa does not publish an annual technology budget. Rajat Taneja said in Oct 2025 that Visa spent about $12 billion on technology over 5 years. Visa said on Oct 27, 2026 that it had invested $15 billion in network security and resilience over the last 5 years. Visa's network and processing expense rose 19% to $1.07 billion in fiscal 2026. The Form 10-K attributes the rise to investment in technology and the processing network and to acquired businesses, and gives no AI share.
What AI itself costs
Visa has not disclosed what it spends on AI each year. Rajat Taneja said in Oct 2025 that Visa spent $3.5 billion over 5 years building its AI platform. In Feb 2025, Taneja described the same figure as the cost of rebuilding Visa's data platform. Sam Hamilton cited about $3 billion spent on Visa's data and AI stack over 10 years (Mar 2025), so the figures overlap rather than add up. Visa told Business Insider that its staff used 1.9 trillion AI tokens in Mar 2026, and Visa has run a $100 million generative AI ventures fund for startups.
Reported ROI
Visa has reported no business-level return that it credits to AI alone. Visa's fraud figures, such as about $40 billion in fraud attempts prevented a year (Feb 2025), count losses avoided by Visa's clients, not by Visa itself. Visa's value-added services revenue grew 30% in constant dollars to about $14 billion in fiscal 2026. Ryan McInerney tied 350+ product releases in fiscal 2026, up 25%, to the increased use of AI. Visa does not split either figure by AI. Visa credits its one reported firm-level cost change, the Jul 2026 job cuts, to an efficiency program in which AI is one factor.
Headcount
Visa announced on Jul 28, 2026 that it would cut about 2,600 jobs, 7% of its workforce, mostly in technology and product teams, and recorded $563 million of severance in its Q3 fiscal 2026 results. Ryan McInerney wrote that AI is helping accelerate the evolution of how work gets done at Visa, and told analysts the savings will be reinvested in growth areas. Bloomberg reported that AI had cut repetitive tasks and sped up product development but was not the only reason for the cuts. Visa's fiscal 2026 Form 10-K credits the severance to actions to drive operational efficiencies and reinvest in high-growth opportunities. The 10-K reports about 34,800 employees at Sep 30, 2026, up from about 34,100 a year earlier, and says personnel expense rose partly because of more employees, including from acquisitions.
Productivity markers
Ryan McInerney said on Jul 28, 2026 that Visa teams on the Agentic Toolchain post 80% more code commits, cut the time to define requirements from 30 days to 5 and develop features 65%+ faster. McInerney said on Oct 27, 2026 that 12 people designed, built and shipped Visa's stablecoin platform support for Open USD in 7 weeks. Chris Suh said AI helped Visa Consulting & Analytics deliver 1,200 projects in the quarter to Jun 2026. Visa has not converted any of these speed or throughput measures into cost or headcount figures.
Governance and risk
Board oversight
Visa's board oversees AI as part of its strategy, reviewing how management uses AI to drive innovation, operational efficiency and customer experience. The board's Audit and Risk Committee oversees AI risks, including data privacy, cybersecurity, regulatory compliance and ethics (Dec 2025 proxy statement). Visa publishes AI Principles in its corporate responsibility report.
Model governance and the AI observatory
Rajat Taneja told Fortune in Oct 2025 that Visa codified its data-use principles into observability systems. Visa's AI observatory dashboard catalogs every model with its compliance, risk, privacy and security profile, and sends real-time alerts when a model drifts from its guardrails. Taneja said in Feb 2025 that Visa set rules after ChatGPT's release to protect its data, avoid copyright infringement and keep its models compliant.
Agents as identities
Visa's Project Glasswing white paper lists treating AI agents as identities among 12 required practices (VentureBeat, Jul 2026). Under that practice, every agent that calls an API or changes a system should have scoped permissions, least privilege and full audit trails, and be included in identity governance. Visa's vulnerability harness enforces deterministic controls, policy gates and human review at each stage, and Visa now makes AI security posture part of supplier due diligence.
Controls on agentic payments
Visa's Trusted Agent Protocol uses agent-specific cryptographic signatures so merchants can tell approved shopping agents from malicious bots (Oct 2025). Visa said agent payments through OpenAI will run within user-set permissions such as spending limits, merchant categories or required approvals, using tokenized credentials and real-time fraud monitoring (Jun 2026).
Disclosed risks and regulation
Visa's fiscal 2026 Form 10-K warns that Visa's AI depends partly on third-party tools, models and infrastructure it does not fully control. The 10-K also flags the EU AI Act, whose rules for most high-risk systems take effect in 2027, and a patchwork of US state laws. The 10-K says agentic commerce raises unsettled questions of liability for unauthorized or erroneous agent transactions.
Recent signals
Nov 5, 2026
Visa's fiscal 2026 Form 10-K described the Visa Agentic Toolchain and an internal generative AI hub and AI assistant for employees, and reported about 34,800 employees at Sep 30, 2026.
Oct 27, 2026
Ryan McInerney said more than 60% of Visa's product development and engineering teams had adopted the agentic way of working, and that nearly 50 AI partners were live on Visa Intelligent Commerce. McInerney said Visa is working with Anthropic to bring Anthropic's commerce blueprints to merchants.
Sep 2026
Anthropic launched a blueprint for building commerce agents on Claude, naming Visa among the partners helping merchants use it.
Aug 3, 2026
Visa agreed to buy BioCatch, a provider of behavioral-first, multi-signal fraud intelligence, for $2.4 billion in cash, with closing expected by the end of its fiscal second quarter of 2027.
Jul 28, 2026
Visa said it would cut about 2,600 jobs, mostly in technology and product teams, booked $563 million of severance, and said AI is helping change how work gets done. Ryan McInerney described agentic squads of 2 to 4 that develop features 65%+ faster.
Jul 2026
Rajat Taneja told VB Transform that Claude Mythos Preview chained minor weaknesses in Visa's network into working exploit paths, which Visa's zero-trust controls stopped. Taneja said Visa now measures security by Mean Time to Adapt.
Jun 10, 2026
Visa announced a partnership with OpenAI at the Visa Payments Forum to bring Visa payments into agentic commerce on OpenAI's platforms. Visa released the Visa Vulnerability Agentic Harness on GitHub the same day.
Apr 2026
Visa told Business Insider that 89% of its employees are active AI users, that token use doubled from Feb to Mar 2026 to 1.9 trillion, and that it rewards teams that use AI to work faster.
Apr 1, 2026
Visa launched 6 AI-powered dispute tools for merchants, issuers and acquirers after a record 106 million disputes in 2025.
Dec 18, 2025
Visa said partners had completed hundreds of agent-initiated transactions on Visa Intelligent Commerce and predicted that millions of consumers would use AI agents to buy by the 2026 holiday season.
Quick answers
- How does Visa deploy AI?
- Visa deploys AI in its own risk models, which score every transaction on VisaNet, and in generative and agentic tools for staff, built on OpenAI, Anthropic and Google models. Visa said in Apr 2026 that 89% of its employees actively use AI, and in Oct 2026 that more than 60% of its product and engineering teams build with its in-house Agentic Toolchain. Visa announced about 2,600 job cuts in Jul 2026, saying AI is helping change how work gets done, but has published no return it credits to AI alone.
- What does Visa use AI for?
- Ryan McInerney said Visa had more than 150 AI-powered applications in Jul 2026. Visa's AI runs in fraud and risk scoring on VisaNet, agentic commerce, dispute handling, consulting, client service, cybersecurity and Visa's own product development. Visa has published no business return it credits to AI for any named use case. Visa's fraud figures count losses prevented for issuers and sellers, and its Agentic Toolchain results are speed and throughput measures. Visa's value-added services revenue rose 30% in constant dollars to about $14 billion in fiscal 2026, but Visa does not split it by AI. At firm level, Visa announced about 2,600 job cuts in Jul 2026 with AI cited as one factor, and booked the $563 million of severance as an efficiency program.
- Who leads AI at Visa?
- AI at Visa is led by Rajat Taneja (President, Technology); Andres Vives (Chief Data Officer); Subra Kumaraswamy (Chief Information Security Officer). Ownership of AI on the org chart: Rajat Taneja, President, Technology and an executive officer.
- What AI platform does Visa use?
- Visa built its own data and AI platform, which Rajat Taneja said cost $3.5 billion over 5 years (Oct 2025). Visa also runs Secure ChatGPT, a multi-model staff interface on Microsoft Azure, and the Visa Agentic Toolchain for building products (fiscal 2026 Form 10-K).
- Which AI models does Visa use?
- Visa staff use Anthropic's Claude most, closely followed by OpenAI's ChatGPT, with Google's Gemini also popular (Apr 2026). Visa also trains its own risk models, including a Large Transaction Model built on billions of transactions.
- How many Visa employees use AI?
- Visa told Business Insider in Apr 2026 that 89% of its employees are active AI users and 44% are power users, averaging at least 25 prompts a day on 15 days a month.
- What return has Visa reported from AI?
- Visa has reported no business return it credits to AI alone, so its stage rests on rollout evidence and sits at Piloting. Visa announced about 2,600 job cuts on Jul 28, 2026, with $563 million of severance, and CEO Ryan McInerney wrote that AI is helping accelerate the change. Visa's fiscal 2026 Form 10-K credits the severance to actions to drive operational efficiencies and reinvest. Bloomberg reported that AI was not the only factor, and Visa's headcount rose to about 34,800 by Sep 30, 2026. Visa's fraud figures count losses prevented for its clients, and its Agentic Toolchain results (80% more code commits, 65%+ faster feature development, Jul 2026) are speed measures. On rollout, Visa told Business Insider in Apr 2026 that 89% of its employees are active AI users and 44% are power users. Rajat Taneja, the executive officer who is President, Technology, owns the platform. Visa falls short of Scaling because it reports no return it credits to AI in any function and no firm-level AI figure.
Open questions
- What cost, headcount or revenue does Visa credit to AI alone, given that it has published speed measures for the Agentic Toolchain but no cost or headcount equivalent?
- How many of the roughly 2,600 roles Visa cut in Jul 2026 did AI make unnecessary, as opposed to cost management and reinvestment? How many were refilled in growth areas, given that Visa's headcount ended fiscal 2026 higher?
- Beyond Microsoft Azure for Secure ChatGPT, which clouds and data centers run Visa's generative AI hub, employee assistant and Agentic Toolchain, and does any generative model run inside Visa's own data centers?
- What does Visa spend on AI each year, including tokens and model licenses, and how does the $3.5 billion platform figure split between data infrastructure and AI?
- How much revenue do Visa's AI-based products, such as Visa Advanced Authorization, Visa Protect, Featurespace and the dispute suite, bring in within the roughly $14 billion of value-added services revenue?
- How many agent-initiated transactions and how much payment volume run through Visa Intelligent Commerce, given that Ryan McInerney called the numbers very small for fiscal 2027?
- Who sets AI policy at Visa among Rajat Taneja's technology organization, Andres Vives's Global Data Office and the data and AI platform team? Who has led the platform team since Sam Hamilton left Visa by Mar 2026, and where does the Chief Data Officer report?
- How many AI engineers does Visa employ after the Jul 2026 cuts, given that the last figure, more than 2,500, dates from Oct 2025?
- What did Visa's own Project Glasswing testing find in numbers, and how many of the issues has it fixed?
- Why do Visa's application counts differ so much, from more than 500 generative AI applications (2024) to more than 100 internal AI business applications (Oct 2025) and more than 150 AI-powered applications (Jul 2026)?
Sources
Visa
- Visa Intelligent Commerce launch at the Global Product Drop (Apr 30, 2025)
- Visa introduces Trusted Agent Protocol (Oct 14, 2025)
- Visa and partners complete secure AI transactions (Dec 18, 2025)
- Visa partners with OpenAI (Jun 10, 2026)
- Visa Advanced Authorisation product page
- Visa appoints Andrew Torre as President of Value-Added Services (Jun 3, 2025)
Filings and investor materials
News
- Business Insider: Visa says it is burning through trillions of AI tokens a month (Apr 2026)
- VentureBeat: Visa used Mythos to hunt for bugs in its own payment network (Jul 2026)
- SiliconANGLE: Visa partners with OpenAI to let AI agents make payments (Jun 10, 2026)
- Fox Business: Visa slashes thousands of jobs in efficiency push (Jul 28, 2026)
- Cyber Daily: Visa to lay off about 2,600 staff (Jul 2026)
- diginomica: Visa CEO's message to staffers on AI and job cuts (Jul 29, 2026)
- Resultsense: Visa launches AI-powered dispute resolution suite (Apr 2, 2026)
- Fortune: How Visa wove AI into every facet of the company (Oct 7, 2025)
- Fortune: How Visa places big bets on AI and gen AI (Feb 5, 2025)
- PYMNTS: Visa AI head predicts a very fast timeline for GenAI (Jan 22, 2025)
- PYMNTS: Visa uses more than 500 generative AI applications (2024)
- CDO Magazine: Andres Vives, Visa Chief Data Officer, joins Global Board (May 2026)
- CDO Magazine: Visa appoints Andres Vives as Chief Data Officer (2024)
- Evanta: Sam Hamilton, Head of Data and AI at Visa
- TechIntelPro: former Visa data and AI head Sam Hamilton named an AI advisor at an agentic payments startup (Mar 18, 2026)
- VentureBeat: Visa's AI edge, RAG-as-a-service and deep learning (Mar 17, 2025)
Data confidence
- Confirmed
- Stated in a primary or reputable public source.
- Estimated
- Derived or approximated from public reporting.
- Unconfirmed
- Reported, but not yet corroborated.
Visa profile · Enterprise AI diffusion report by Imagine · facts checked against their sources on Nov 9, 2026
