OpenAI Pitches Frontier AI as Infrastructure for Financial Services

Katy Elkin frames OpenAI’s financial-services strategy around three questions she says recur in conversations with banks, insurers, asset managers, market-infrastructure companies, and other financial institutions: how AI can change the workforce, how workflows can better capture AI’s impact, and how AI can be embedded into products in ways that help firms compete.

The organizing claim is that financial institutions should not treat AI as a point solution layered onto existing processes. Elkin argues that the technology now permits firms to redesign workflows “from the ground up,” particularly in areas long constrained by cost and complexity, including risk, compliance, operations, fraud detection, mortgage processing, customer service, digital banking, and back-office work.

OpenAI’s stated model for the sector has three pillars: drive value with AI workflows, multiply workforce impact, and win through AI-powered products. The examples Elkin gives are all positioned inside that framework rather than as isolated deployments.

NatWest is presented as an example of workflow redesign. Elkin says the bank is partnering with OpenAI to help customers with “everyday financial challenges,” including mortgages, banking, and money management. She says NatWest is exploring more than 200 AI projects, with more than 25 already live in production, and that Cora+, its AI assistant, has helped raise customer satisfaction by more than 150%.

Commonwealth Bank of Australia is used to make the workforce-productivity case. Elkin describes CBA as Australia’s largest bank and says it has rolled out ChatGPT Enterprise to 50,000 employees. The goal, in her description, is not just access to a chatbot but “AI fluency” at firm scale: helping teams use ChatGPT, apps, and custom agents in everyday workflows across customer service, fraud, digital banking, and operations. She also says CBA is bringing a cyber initiative to more than 1 million small businesses in partnership with OpenAI, making it, in her view, an example of a regulated bank rolling out AI firm-wide.

Revolut is the product example. Elkin says the company is working with OpenAI to embed AI across the customer experience, including AI-powered financial-crime detection and Rita, its AI assistant. She describes the intended result as stronger fraud detection, less operational friction, and faster support for millions of customers.

The most explicit product-side claim is that AI changes the economics of customization. Elkin says the level of customization once reserved for high-net-worth clients can now be scaled across the customer base. In financial services, that points toward AI-assisted or AI-native products for wealth management, support, compliance, fraud, and advisory-style interactions.

The presentation lists additional active or prospective initiatives OpenAI says it is working on with financial-services customers in Europe and globally: AI research for trading, real-time AI fraud detection, personalized wealth management, intelligent back-office operations, modernizing code infrastructure, and “an AI teammate for every analyst.” Elkin emphasizes that these are “real initiatives” being tackled with customers, not merely abstract use cases.

Her examples also show the breadth of OpenAI’s intended surface area inside financial institutions. The same platform is being pitched for customer-facing assistants, financial-crime tooling, employee productivity, software modernization, research workflows, operations, and fraud response. The underlying commercial argument is that a firm should develop the capacity to absorb model improvements continuously, rather than evaluate each use case as a one-off automation project.

Katy Elkin argues that staying near the frontier matters because AI is advancing unusually quickly and “there are no signs of it slowing down.” A timeline shown alongside the presentation traces OpenAI’s progression from its founding in 2015 through API releases, GPT-1, GPT-2, GPT-3, ChatGPT, ChatGPT Enterprise, GPT-4, Sora, GPT Store, o1-preview, GPT-5, and projected GPT-5.x releases through GPT-5.5, with “Today” marked on the curve.

The point of the timeline is competitive rather than historical: keeping up with AI is presented as a capability in itself. OpenAI’s claim is that financial-services firms need an operating model that can take advantage of model improvements as they arrive.

GPT-5.5 is described as OpenAI’s “most capable model yet,” designed for professional work such as financial analysis and writing code. Elkin points to GDPval, an OpenAI evaluation “specifically designed to test models for their ability to complete economically valuable work,” including spreadsheets, presentations, and research. A chart shown in the presentation compares models on “Rate vs industry professional” and reports GPT-5.5 at the top, with 84.9% wins and 68.2% ties, compared with GPT-5.4, GPT-5, Claude Opus 4.7, Gemini 3.1 Pro, and an industry-expert baseline.

Elkin also says OpenAI has worked to make GPT-5.5 more intuitive and natural to interact with, while making it more efficient and reducing cost with each generation. A displayed post from Greg Brockman characterized GPT-5.5 as “a new class of intelligence” that can complete challenging tasks with little micromanagement, run with low latency at scale, and move toward “a new way of getting computer work done.” The post is used to support the same theme: less supervision, more task completion, and lower operational friction.

Security is not presented as a side concern. Elkin says that as models become more intelligent and capable of advanced coding and computer use, financial-services leaders are increasingly concerned about cybersecurity. OpenAI’s answer, in the presentation, is Codex Security.

She says OpenAI launched Codex Security two months earlier as a tool to identify and patch software vulnerabilities directly in a customer’s codebase. The screenshots shown illustrate a repository scan, a generated threat model, evidence of a vulnerability, and a proposed code fix. One displayed example identifies refund tokens inside links as base64-encoded JSON without a signature, meaning clients could forge or tamper with claims. A later view says the vulnerability was confirmed on HEAD and fixed by adding refund-token signing and verification helpers using HMAC-SHA256 and timing-safe signature comparison.

The demonstration matters because it positions coding agents not only as productivity tools but as controls tooling: scanning repositories, reasoning about threat models, surfacing exploit paths, and producing patches. In the example shown, the system moves from detection to remediation, including code diffs and a pull-request view indicating that merging can be performed automatically.

Elkin also announces a “trusted access program” for trusted enterprise customers to access a special GPT-5.5 cyber model. The stated purpose is to let those customers red-team and catch vulnerabilities in infrastructure and applications before malicious actors can exploit them.

OpenAI’s financial-services pitch ends on governance, residency, and enterprise control. Elkin says the platform runs with encryption in transit and at rest, support for enterprise key management, and data-residency support across Australia, Canada, Europe including the EEA and Switzerland, India, Japan, Singapore, South Korea, the United Arab Emirates, the United Kingdom, and the United States.

The new announcement for the audience is inference residency in Europe. Elkin says that, for the first time, OpenAI is announcing European inference residency, “ensuring that the GPUs powering all of your intelligence reside directly in Europe.” For European financial institutions, the claim is not only that stored data can remain in designated regions, but that model inference can also be run on European infrastructure.

The broader platform vision is described as helping firms govern, monitor, and scale AI across workforce, workflows, and products. Elkin says the event will show that through live demos, customer examples, and OpenAI’s own finance team using the company’s platform and technologies. The final message is that OpenAI wants to build these solutions with financial institutions rather than merely sell access to models.