OpenAI Launches GPT-5.5 Cyber Model to Pitch Banks on AI Integration

OpenAI

OpenAI is pitching financial services firms on rebuilding core workflows and deploying custom AI agents to boost productivity, cut operating friction, and drive hyper-personalized customer experiences.

AI-assisted summary of Frontier Intelligence & Financial Services: Katy Elkin, GTM Lead, OpenAI — original source. Check the original for full context.

Source published: 2026-07-03

Detailed notes

## Background Published by OpenAI on YouTube on June 08, 2026, this video features Katy Elkin presenting OpenAI's vision for the financial services industry. The page description indicates that the presentation covers three main pillars: transforming workflows, amplifying workforce productivity, and building AI-native customer products.

## TL;DR Katy Elkin outlines how top financial institutions are deploying OpenAI's technology to revolutionize operations, boost productivity, and power customer-facing products. She highlights real-world enterprise deployments, including NatWest's Cora Plus assistant, Commonwealth Bank of Australia's firm-wide rollout of ChatGPT Enterprise to 50,000 employees, and Revolut's AI-driven financial crime detection. Additionally, she introduces GPT-5.5 alongside new enterprise security features like the Trusted Access Program and European GPU inference residency to meet strict compliance standards.

## Atmosphere & style The presentation carries an authoritative, polished, and forward-looking corporate vibe typical of high-level enterprise keynote addresses. Elkin speaks with executive warmth and professional enthusiasm, blending strategic industry insights with personal reflections on her early career as a business analyst. The narrative moves seamlessly from practical institutional case studies to cutting-edge model announcements, instilling a sense of rapid technological momentum and enterprise-grade trust.

## Key takeaways - [00:00](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=0s) Financial services leaders consistently focus on three core strategic questions: optimizing the workforce, re-engineering workflows, and infusing AI into customer products. - [00:00](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=0s) Rebuilding risk, compliance, and back-office operations from the ground up enables institutions to overcome historical cost and operational complexity constraints. - [00:00](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=0s) NatWest partnered with OpenAI to address everyday customer financial challenges, exploring over 200 AI projects with more than 25 deployed in production. - [00:00](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=0s) NatWest's AI assistant, Cora Plus, achieved a customer satisfaction score increase of over 150%. - [01:40](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=100s) AI acts as an 'always-on coworker' by automating tactical execution, allowing employees to focus on revenue generation and client relationships. - [01:40](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=100s) Commonwealth Bank of Australia (CBA) rolled out ChatGPT Enterprise to 50,000 employees to drive firm-wide AI fluency across core banking operations. - [01:40](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=100s) CBA and OpenAI collaborated on a cybersecurity initiative impacting over 1 million small businesses. - [01:40](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=100s) AI allows financial institutions to scale hyper-personalized services, previously reserved for high-net-worth clients, to retail banking customers. - [01:40](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=100s) Revolut integrates OpenAI models across operations, from AI-powered financial crime detection to its customer support assistant, Rita. - [01:40](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=100s) OpenAI actively collaborates with financial institutions on AI-driven trading research, personalized wealth management, and legacy COBOL system modernization. - [03:48](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=228s) GPT-5.5 is optimized for complex professional tasks like financial analysis and software development, offering higher speed and lower token costs. - [03:48](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=228s) The GDPval benchmark measures a model's ability to execute economically valuable work such as financial research, spreadsheets, and presentations. - [03:48](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=228s) OpenAI launched the Trusted Access Program, granting enterprises access to a specialized GPT-5.5 cyber model for proactive red-teaming and vulnerability discovery. - [03:48](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=228s) To satisfy strict European regulatory standards, OpenAI announced European inference residency alongside existing regional data storage residency.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=0s) Three Core Questions and Transforming Financial Workflows

Katy Elkin, Go-To-Market (GTM) Lead at OpenAI, opens her address by highlighting the rare privilege of gathering an in-person room full of top financial services executives. Reflecting on her regular conversations with leading financial firms, she observes that no matter the institution, leaders consistently ask the same three core strategic questions: how to rethink their workforce to maximize AI value, how to re-engineer workflows to better capture AI's impact, and how to infuse AI directly into products to maintain a competitive market edge.

To illustrate how institutions are successfully transforming workflows, Elkin points to processes in risk, compliance, and back-office operations—areas that have historically been constrained by massive costs and operational complexity. By rebuilding these processes from the ground up, banks can change how they operate fundamentally. A prime example is UK bank NatWest, which has partnered with OpenAI to help customers manage everyday financial challenges ranging from mortgages to general money management.

- **NatWest AI Traction:** NatWest is actively exploring over 200 AI projects, with more than 25 already live in production. - **Customer Satisfaction Impact:** NatWest’s AI-powered assistant, Cora Plus, has driven a customer satisfaction score increase of over 150%.

[01:40](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=100s) Amplifying Workforce Productivity and Redefining Customer Products

Moving to the second pillar—workforce amplification—Elkin describes AI as an "always-on coworker." Autonomous agents handle tactical execution and surface key insights, freeing top talent to concentrate on building client relationships and driving revenue. She highlights Commonwealth Bank of Australia (CBA), Australia's largest bank, which deployed ChatGPT Enterprise across 50,000 employees. CBA's deployment centers on driving firm-wide "AI fluency," integrating custom agents and apps into everyday customer service, fraud prevention, digital banking, and operations.

On the third pillar—product innovation—Elkin explains that AI enables institutions to scale bespoke experiences, offering hyper-personalized services once reserved strictly for high-net-worth clients to their broader retail customer base. She highlights digital bank Revolut, which embeds OpenAI’s technology across its operations—from AI-powered financial crime detection to its customer-facing AI assistant, Rita. Elkin adds that OpenAI is actively collaborating with global institutions on initiatives such as AI-driven trading research, personalized wealth management, and modernizing legacy mainframe COBOL systems.

- **CBA Scale & Outreach:** CBA deployed ChatGPT Enterprise to 50,000 workers and partnered with OpenAI on a cybersecurity initiative reaching over 1 million small businesses. - **Revolut Results:** Embeds AI into its assistant Rita and financial crime detection systems, reducing operational friction and accelerating support for millions of users.

[03:48](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=228s) Frontier Models, GDPval Benchmark, and Enterprise Security Innovations

Elkin emphasizes that maintaining an edge is more crucial than ever as AI capabilities accelerate faster than any previous technology. She introduces GPT-5.5 as OpenAI's most capable model yet, specifically optimized for complex professional tasks such as financial analysis and software development. She references "GDPval"—an evaluation benchmark designed to test a model's ability to complete economically valuable work like spreadsheets, research, and presentations—which demonstrates GPT-5.5's state-of-the-art performance. Recalling her own early career as a business analyst, Elkin notes that these benchmark capabilities are "downright thrilling."

Addressing the critical cybersecurity concerns of financial executives, Elkin outlines OpenAI’s latest enterprise defense updates. Following the release of Codex Security two months prior to identify and patch code vulnerabilities, OpenAI is officially launching its Trusted Access Program. This program gives enterprise customers access to a specialized GPT-5.5 cyber model for red-teaming applications and catching system vulnerabilities before malicious actors can exploit them. Furthermore, to satisfy strict regulatory requirements for European institutions, OpenAI is announcing European inference residency alongside existing data storage residency, ensuring that the GPUs powering enterprise intelligence physically reside within Europe.

- **GPT-5.5 & GDPval:** GPT-5.5 delivers state-of-the-art performance on economically valuable tasks, pairing higher speed and lower token costs with natural, intuitive interactions. - **Cybersecurity Tools:** Features Codex Security alongside the new Trusted Access Program for the GPT-5.5 cyber model to enable proactive red-teaming. - **Data Protection & European Residency:** Includes end-to-end encryption, Enterprise Key Management (EKM), data storage residency, and newly announced in-region European GPU inference residency.

## Notable quotes > [01:40](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=100s) always-on coworker — Katy Elkin

> [01:40](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=100s) AI fluency — Katy Elkin

> [03:48](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=228s) downright thrilling — Katy Elkin

## Mentioned Frontier Intelligence & Financial Services, Katy Elkin, OpenAI, NatWest, Cora Plus, Commonwealth Bank of Australia, ChatGPT Enterprise, Revolut, Rita, COBOL, GPT-5.5, GDPval, Codex Security, Trusted Access Program, Enterprise Key Management

[Source](https://www.youtube.com/watch?v=ykNmTvRcqPQ)

中文简报

OpenAI推出GPT-5.5 Cyber模型，力推银行业整合AI OpenAI正力推金融服务机构重构核心工作流程并部署定制AI Agent，以提升生产力、减少运营阻力并推动高度个性化的客户体验。 Katy Elkin阐述了顶尖金融机构如何部署OpenAI的技术以彻底变革业务运营、提升生产力并赋能面向客户的产品。她重点介绍了多个实际的企业级部署案例，包括NatWest的Cora Plus助手、澳大利亚联邦银行（Commonwealth Bank of Australia）向5万名员工全员推广ChatGPT Enterprise，以及Revolut基于AI的金融犯罪检测。此外，她还介绍了GPT-5.5以及可信访问计划（Trusted Access Program）和欧洲GPU推理数据驻留等全新企业安全特性，以满足严苛的合规标准。金融服务业领袖始终聚焦于三个核心战略问题：优化员工队伍、重构工作流程以及将AI注入客户产品。 (0:00) 从底层全面重构风险、合规与后台运营，使金融机构得以克服以往在成本和运营复杂性方面的制约。 (0:00) NatWest与OpenAI合作应对日常客户财务难题，共探索了200多个AI项目，其中超过25个已在生产环境中部署。 (0:00) NatWest的AI助手Cora Plus使客户满意度评分提升了150%以上。 (0:00) AI通过自动化战术执行扮演“全天候同事”的角色，让员工能够专注于创造收入和维护客户关系。 (1:40)

Reading summary

OpenAI Go-To-Market Lead Katy Elkin asserts that financial institutions are navigating competitive pressures by deploying frontier AI models to rebuild legacy workflows, amplify employee capacity, and scale personalized financial products. Re-engineering Workflows in Core Banking Operations (0:00) According to Elkin, financial executives routinely focus on how to re-engineer workflows, rethink organizational talent, and embed AI directly into commercial offerings. Rather than applying incremental updates, institutions are targeting historically cost-intensive and complex segments like compliance, risk management, and back-office administration. As an example of operational restructuring, UK bank NatWest has investigated more than 200 AI projects, deploying over 25 into production. These initiatives include its customer-facing assistant, Cora Plus, which NatWest reports has delivered a customer satisfaction score increase exceeding 150% across everyday banking workflows. Workforce Fluency and Scaling Bespoke Products (1:40) Elkin frames AI agents as continuous collaborators that manage tactical execution, permitting personnel to concentrate on revenue generation and client advisory. Commonwealth Bank of Australia (CBA) exemplifies this scale by deploying ChatGPT Enterprise across 50,000 employees to cultivate enterprise-wide AI fluency across fraud prevention, digital banking, and operational pipelines, while also delivering a cybersecurity program reaching over 1 million small businesses. On the product front, institutions use AI to extend bespoke advisory services—previously limited to high-net-worth clients—to mainstream retail accounts. Digital bank Revolut has integrated OpenAI models into its customer support agent Rita and its financial crime monitoring systems. Additionally, firms are collaborating with OpenAI to modernize legacy mainframe COBOL code, conduct AI-assisted trading research, and deliver personalized wealth management. Frontier Model Benchmarks and Security Safeguards (3:48) To address professional demands, Elkin highlights GPT-5.5, which is evaluated on OpenAI's 'GDPval' benchmark measuring economically valuable capabilities such as financial analysis, spreadsheet modeling, research, and presentation generation. Elkin claims the model provides state-of-the-art performance with lower token costs and faster response times. Addressing institutional risk and regulatory constraints, OpenAI introduced its Trusted Access Program alongside Codex Security, giving enterprises access to a specialized GPT-5.5 cyber model for proactive system red-teaming. For European financial institutions subject to strict data governance, OpenAI announced in-region European GPU inference residency to complement existing data storage residency, supported by end-to-end encryption and Enterprise Key Management. Elkin emphasizes that maintaining a competitive edge requires financial institutions to balance advanced model performance on economic tasks with rigorous compliance, proactive red-teaming, and localized infrastructure controls.

精炼阅读摘要

OpenAI全球业务拓展负责人Katy Elkin指出，金融机构在布局人工智能时主要聚焦于重塑业务流程、倍增员工生产力以及打造智能化产品三大核心战略；为此，OpenAI正通过前沿模型GPT-5.5、专项网络安全防护方案及欧洲本地化GPU推理驻留，支持金融机构平衡业务创新与合规安全。重塑后台流程与普惠化定制产品 (0:00) 传统金融机构的风控、合规与后台运营长期面临高昂成本与流程复杂度的制约。Elkin指出，通过自底向上重构业务流，金融机构能够从根本上转变运营模式。以英国国民西敏寺银行（NatWest）为例，该行正推进超过200个AI项目，其中逾25个已正式上线运行；其基于OpenAI技术打造的客户助手Cora Plus使客户满意度评分提升了150%以上。在客户端产品层面，AI使机构得以将过去仅面向高净值客户的超个性化定制服务，规模化推广至普通零售客户。数字银行Revolut已将OpenAI技术全面嵌入运营，包括客户智能助手Rita以及金融犯罪监测系统，以减少业务摩擦。此外，OpenAI亦与全球金融机构展开合作，探索AI驱动的交易研究、个性化财富管理以及传统主机COBOL老旧系统的现代化改造。释放员工产能与构建全员AI素养 (1:40) 针对劳动力维度的赋能，Elkin将AI定义为金融从业者的“全天候协同办公伙伴”。自主智能体负责承担具体执行与提炼关键洞见，从而释放核心业务人员的时间与精力，使其专注于建立客户信任与创造业务营收。澳大利亚联邦银行（CBA）是全员赋能的代表案例，该行已面向50,000名员工全面部署ChatGPT Enterprise，核心目标在于建立全行范围的“AI素养”。CBA将定制智能体与应用程序融入日常客户服务、反欺诈、数字银行及中后台运营中，同时与OpenAI合作开展网络安全行动，服务范围覆盖超过100万家小型企业。 GPT-5.5能力突破与GDPval经济价值评估 (3:48) 随着AI技术演进速度超越以往任何技术浪潮，金融机构保持竞争壁垒的需求愈加迫切。Elkin介绍了OpenAI推出的前沿模型GPT-5.5，该模型针对金融分析、软件开发等复杂专业任务进行了专项优化，在提升处理速度的同时降低了Token使用成本，并提供更为自然的交互体验。为衡量模型执行高价值经济任务的实际能力，行业引入了“GDPval”评估基准，专门测试模型在电子表格构建、专业研究和演示文稿制作等场景下的表现。Elkin结合自身早期的商业分析师从业经历表示，GPT-5.5在GDPval测试中展示的顶尖性能，印证了其承担复杂专业工作的实用价值。企业级主动防御体系与欧洲合规驻留 (3:48) 针对金融机构高管最关切的网络与信息安全问题，OpenAI拓展了企业级防御产品矩阵。继两个月前发布用于识别与修补代码缺陷的Codex Security之后，OpenAI正式推出“可信访问计划”（Trusted Access Program），为企业客户提供专用的GPT-5.5网络安全模型，便于企业在漏洞被恶意利用之前主动开展红队测试与系统排查。为满足欧洲机构严苛的监管要求，OpenAI除提供端到端加密与企业密钥管理（EKM）外，在既有的欧洲数据存储驻留基础上，正式宣布支持欧洲区域内的物理GPU推理驻留，确保驱动企业AI运转的算力硬件实际位于欧洲境内。 Elkin的阐述表明，前沿AI在金融领域的落地正在从外围探索迈向核心生产系统；机构要在激烈的市场竞争中确立优势，不仅需要检验模型处理专业经济任务的效能，更必须依托严密的红队攻防工具与本土化硬件驻留方案，以满足金融监管与数据合规的底线要求。

Detailed notes — zh

## Background 该视频由 OpenAI 官方于 2026 年 7 月 3 日发布在 YouTube 平台。视频简介指出，内容主要展示了 OpenAI 关于金融服务机构如何围绕三大支柱（变革工作流、倍增劳动力生产力以及构建原生 AI 客户产品）应用人工智能的战略愿景。 ## TL;DR OpenAI 金融服务 GTM 负责人 Katy Elkin 面向行业高管系统阐述了金融机构拥抱 AI 的三大支柱：重构复杂业务工作流、打造全天候数字同事以提升全员劳动力效能，以及重塑面向客户的普惠金融产品。演讲结合了英国国民威斯敏斯特银行（NatWest）、澳大利亚联邦银行（CBA）和 Revolut 等头部机构的实战案例，并展示了专为高价值知识工作打造的 GPT-5.5 模型及 GDP-Val 基准表现。为化解严苛的金融监管与安全疑虑，OpenAI 正式推出用于红队演练的 GPT-5.5 Cyber 专用模型，并首次宣布在欧洲实现算力底层的物理“推理数据本地化”。 ## Atmosphere & style 整场演讲干练沉稳且高度契合行业脉搏，主讲人结合自身早年商业分析师的从业背景与现场高管迅速建立起信任共鸣。叙事节奏紧凑洗练，在对 COBOL 大型机等传统金融痛点的风趣调侃中，自然过渡到严肃的架构安全与合规治理探讨。随着一系列重磅技术与本地化基建举措的宣布，整场演说散发出前沿科技深度扎根于实体金融核心业务的务实感与突破力。 ## Bulletin - [00:00](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=0s) OpenAI 提出金融业落地 AI 的三大核心支柱：工作流变革（Workflows）、劳动力倍增（Workforce）以及客户产品重塑（Products）。 - [00:55](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=55s) NatWest 已上线 25 个以上 AI 项目，智能助理 Cora+ 使 CSAT 提升超 150%；澳大利亚联邦银行则向 5 万名员工全面部署 ChatGPT Enterprise。 - [03:48](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=228s) OpenAI 推出最强模型 GPT-5.5，在衡量实际经济价值工作（如财务建模、行研与代码）的 GDP-Val 评估中达到业界顶尖水平。 - [05:07](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=307s) OpenAI 宣布面向企业启动受信任访问计划并开放 GPT-5.5 Cyber 专用模型，同时在欧洲正式落地推理数据本地化（Inference residency）。 ## Key takeaways - [00:00](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=0s) 金融机构决策者普遍面临三大核心议题：如何重塑劳动力队伍、如何重新设计业务工作流，以及如何将 AI 深度融入产品以构建壁垒。 - [00:55](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=55s) 通过重构高成本与高复杂度的风控合规流程，AI 正在从底层革新银行的核心运营模式。 - [00:55](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=55s) NatWest 探索了超 200 个 AI 项目且已有超 25 个投产，AI 助手 Cora+ 推动客户满意度（CSAT）增长超过 150%。 - [01:55](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=115s) AI 正在演变为金融员工“全天候在线的数字同事”，自主接管繁琐事务以释放专业人才聚焦高价值客户关系与营收。 - [02:15](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=135s) 澳大利亚联邦银行（CBA）面向 5 万名全员普及 ChatGPT Enterprise，全面培养员工的“AI 流畅度”，并联合赋能超百万家小微企业网络安全。 - [02:43](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=163s) AI 使得传统仅属于高净值客户的专属顾问服务得以低边际成本平铺普及给广大普通大众。 - [02:43](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=163s) Revolut 将 AI 深度嵌入金融犯罪检测系统及客户助手 Rita，在强化反洗钱防欺诈的同时大幅削减运营摩擦。 - [03:15](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=195s) AI 正攻坚金融机构传统顽疾场景，包括大型机与 COBOL 遗留系统代码现代化重构、投研辅助以及个性化财富管理。 - [03:48](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=228s) GPT-5.5 专为高难度专业知识工作打造，在评估实际商业产出价值的 GDP-Val 基准测试中树立了新行业标杆。 - [04:30](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=270s) 模型代际迭代在实现逻辑推理与计算机操控能力突破的同时，持续降低了单位推理调用成本。 - [05:07](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=307s) 推出代码安全审计工具 Codex Security 与 GPT-5.5 Cyber 专用模型，协助金融企业开展红蓝对抗并主动修复漏洞。 - [05:35](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=335s) 欧洲正式上线“推理数据本地化（Inference residency）”，确保执行 AI 计算的 GPU 算力完全位于欧洲境内，满足严格的合规与数据主权诉求。 - [05:48](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=348s) 提供端到端企业安全保障体系，包括静态/传输全链路加密与企业密钥管理（EKM），构建高安全水准的企业级智能平台。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=0s) 金融机构的核心三问与三大支柱 Katy Elkin 登台向现场的高管们致谢，并提到像今天这样能够将众多金融机构的决策者齐聚一堂实属难得。作为 OpenAI 金融服务团队的 Go-To-Market 负责人，她在与全球顶尖金融机构的日常合作与交流中发现，几乎所有企业都在反复思考并提出相同的三大核心问题：首先，如何重塑劳动力（Workforce），以便让 AI 创造更大的价值？其次，业务工作流（Workflows）该如何重新设计，才能更好地承接并放大 AI 的影响力？最后，如何将 AI 深度注入产品（Products）中，从而击败竞争对手并在市场中立于不败之地？针对这三大支柱，OpenAI 正在与全球领先的金融机构展开深度实践。 ```mermaid flowchart TD accTitle: AI 在金融服务业的三大赋能支柱 accDescr: 展示了从工作流重塑、劳动力赋能到原生产品构建的三大核心支柱及其实践案例。 subgraph Pillars ["金融服务 AI 落地三大支柱"] W1["工作流变革 (Workflows)"] W2["劳动力倍增 (Workforce)"] W3["客户产品重塑 (Products)"] end subgraph Cases ["标杆落地案例"] C1["NatWest Cora+ 智能助理 CSAT 提升 >150%"] C2["澳大利亚联邦银行 (CBA) 全行 5 万人部署 ChatGPT Enterprise"] C3["Revolut Rita 智能助手 & AI 反金融犯罪"] end W1 --> C1 W2 --> C2 W3 --> C3 ``` ### [00:55](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=55s) 标杆实践：从工作流变革到全员劳动力倍增在工作流层面，风控、合规与日常运营等核心环节长久以来受制于高昂的成本和极高的业务复杂度。如今，AI 使得银行能够从底层彻底重新设计自身的工作流程，实现前所未有的运营方式。英国国民威斯敏斯特银行（NatWest）便是一个典型案例，他们与 OpenAI 携手帮助客户解决从抵押贷款、日常银行服务到资金管理等各类日常财务挑战。目前 NatWest 已探索了 200 多个 AI 项目，其中超过 25 个项目已直接上线投入生产；其实装的 AI 助手 Cora+ 更是将客户满意度（CSAT）提升了 150% 以上。在赋能劳动力方面，AI 在金融机构内部正演变为一位“全天候在线的数字同事（Always-on coworker）”。AI Agent 能够自主执行繁琐任务、提炼关键洞察并成倍提升工作产出，从而将企业最核心的专业人才从案头工作中解放出来，专注于维系客户关系和推动营收增长。澳大利亚体量最大的银行——澳大利亚联邦银行（Commonwealth Bank of Australia, CBA）正在践行这一模式。他们向全行 50,000 名员工全面推广了 ChatGPT Enterprise，核心目标在于全面普及全员的“AI 流畅度（AI Fluency）”，让员工在客户服务、反欺诈、数字银行及中后台运营等日常业务中广泛使用 ChatGPT、应用插件及定制化 Agent；与此同时，CBA 还与 OpenAI 合作将网络安全倡议带给了超过 100 万家小微企业，树立了大型强监管银行全行拥抱 AI 的标杆。 - **NatWest 业务突破**：探索超过 200 个 AI 项目，25 个以上已正式上线，智能助手 Cora+ 带动客户满意度评分提高超 150%。 - **CBA 全员普及**：面向 50,000 名员工部署 ChatGPT Enterprise，并携手为 100 多万家小型企业提供网络安全赋能。 ### [02:43](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=163s) 产品重塑与行业前沿场景落地第三大支柱在于产品重塑。AI 正在颠覆金融机构提供服务、顾问咨询以及赢得市场的方式。过去只有高净值客户（High-net-worth clients）才能享受的高度定制化金融服务，现在可以通过 AI 规模化平铺给所有普通客户。英国数字银行独角兽 Revolut 便深度结合 OpenAI 技术，将 AI 全面融入客户体验：一方面构建 AI 驱动的金融犯罪检测体系，另一方面推出了面向用户的 AI 智能助手 Rita。这不仅带来了更强大的欺诈防范能力，还大幅降低了运营摩擦，为数以百万计的客户提供了更即时高效的响应。除了客户服务外，OpenAI 还在欧洲及全球范围内与金融机构探索更加深入的技术与业务课题。Katy 现场调侃并询问台下是否还有人依然在维护老旧的“大型机与 COBOL 遗留系统（Mainframe COBOL legacy）”，并指出诸如交易领域的 AI 投研辅助、底层代码架构的现代化改造、个性化财富管理等场景，都已不再是空谈，而是正在与客户并肩攻坚的实战项目。 ### [03:48](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=228s) GPT-5.5、GDP-Val 与专业级金融能力跃迁技术的迭代节奏前所未有，而一切业务变革的源头都在于底层模型。OpenAI 推出了迄今为止能力最强的模型——GPT-5.5，该模型专为金融分析和专业代码编写等复杂知识工作而设计。在衡量模型完成“具备实际经济价值工作”（例如复杂的电子表格处理、制作演示文稿和行业研究等）的基准测试 GDP-Val 中，GPT-5.5 展现出了无可争议的业界领先水平（State-of-the-art）。Katy 笑着回忆自己早年的职业生涯就是从商业分析师（Business Analyst）起步的，看到 AI 在这些专业领域的惊人表现，感到由衷的激动与震撼。在性能跨越式提升的同时，团队还致力于让模型的交互方式更直观、更自然，并大幅提升执行效率，帮助用户快速出成果；更关键的是，每一代新模型的推出都在持续压低推理成本。随着模型逻辑智能以及高级代码生成和“计算机操控（Computer use）”能力的爆发，金融业高管对网络安全的关切也达到了前所未有的高度。 - **针对性评测指标 GDP-Val**：专门评估模型完成高价值实际任务的能力，涵盖财务电子表格计算、PPT 制作与商业行研。 - **降本增效曲线**：在综合性能不断刷新的同时，代际推理调用成本持续下降。 ### [05:07](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=307s) 企业级网络安全守护与欧洲本地化合规突破为应对严峻的安全挑战，OpenAI 在两个月前推出了代码安全审计工具 Codex Security，协助企业直接在软件中识别并修补代码漏洞。在此基础上，Katy 在大会现场正式宣布启动“受信任访问计划（Trusted Access Program）”：所有受信任的企业级客户均可申请使用专门的“GPT-5.5 Cyber 模型”。企业可以借助该专用模型对自身的基础设施与应用程序开展红蓝对抗（Red teaming），在黑客与恶意攻击者趁虚而入之前提前发现并堵塞系统漏洞。针对对合规和控制力要求极高的金融企业，整套体系均运行在世界级的企业安全控制框架之下。所有数据在传输过程（In transit）与静态存储（At rest）状态下均受到强加密保护，并全面支持企业密钥管理（EKM）。对于欧洲金融机构而言最为关键的是，OpenAI 提供了广泛的数据本地化存储支持（Data residency）；更为重磅的是，Katy 在演讲中首次公开宣布在欧洲上线“推理数据本地化（Inference residency）”，彻底确保支撑各项 AI 计算的物理 GPU 算力均直接部署并留在欧洲境内。 ```mermaid flowchart LR accTitle: OpenAI 企业级安全与合规架构 accDescr: 展示了包含数据加密、专属安全模型、数据存储本地化及欧洲推理算力本地化的企业级安全体系。 subgraph Security ["主动防御与代码安全"] A["Codex Security 自动漏洞检测与修复"] B["GPT-5.5 Cyber 专用模型 红队攻防模拟 / 提前排雷"] end subgraph Enterprise ["合规与主权保障"] C["数据全面加密 (传输/静态) + EKM 密钥管理"] D["数据存储本地化 (Data Residency)"] E["欧洲推理本地化 (Inference Residency / 本地 GPU)"] end Security --> Enterprise ``` - **GPT-5.5 Cyber 模型**：通过受信任访问计划（Trusted Access Program）向企业客户开放，专门用于红队测试与前置防范基础设施漏洞。 - **欧洲推理本地化**：首次实现推理算力本土化部署，驱动智能推理的 GPU 物理硬件完全位于欧洲境内，充分满足严苛的欧洲金融监管与隐私标准。 - **全流程防护机制**：支持传输与静态加密，提供企业级密钥管理（Enterprise Key Management, EKM）能力。在演讲尾声，Katy 总结道，OpenAI 的终极愿景是依托企业级智能平台，协助机构全面管理、监控并规模化拓展 AI 价值。随后的环节不仅会安排丰富的实机演示和客户经验分享，甚至还将现场展示 OpenAI 自身的财务团队是如何在日常业务中使用自家的平台和前沿技术的。 ## Notable quotes > [00:00](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=0s) 首先，如何重塑劳动力，以便让 AI 创造更大的价值？其次，业务工作流该如何重新设计，才能更好地承接并放大 AI 的影响力？最后，如何将 AI 深度注入产品中，从而击败竞争对手并在市场中立于不败之地？ — Katy Elkin > [00:55](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=55s) 风控、合规与日常运营等核心环节长久以来受制于高昂的成本和极高的业务复杂度。如今，AI 使得银行能够从底层彻底重新设计自身的工作流程，实现前所未有的运营方式。 — Katy Elkin > [00:55](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=55s) AI 在金融机构内部正演变为一位“全天候在线的数字同事”。AI Agent 能够自主执行繁琐任务、提炼关键洞察并成倍提升工作产出，从而将企业最核心的专业人才从案头工作中解放出来，专注于维系客户关系和推动营收增长。 — Katy Elkin > [00:55](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=55s) 核心目标在于全面普及全员的“AI 流畅度”，让员工在客户服务、反欺诈、数字银行及中后台运营等日常业务中广泛使用 ChatGPT、应用插件及定制化 Agent。 — Katy Elkin > [02:43](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=163s) 过去只有高净值客户才能享受的高度定制化金融服务，现在可以通过 AI 规模化平铺给所有普通客户。 — Katy Elkin > [02:43](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=163s) 诸如交易领域的 AI 投研辅助、底层代码架构的现代化改造、个性化财富管理等场景，都已不再是空谈，而是正在与客户并肩攻坚的实战项目。 — Katy Elkin > [03:48](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=228s) OpenAI 推出了迄今为止能力最强的模型——GPT-5.5，该模型专为金融分析和专业代码编写等复杂知识工作而设计。 — Katy Elkin > [03:48](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=228s) 在衡量模型完成“具备实际经济价值工作”的基准测试 GDP-Val 中，GPT-5.5 展现出了无可争议的业界领先水平。 — Katy Elkin > [05:07](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=307s) 所有受信任的企业级客户均可申请使用专门的“GPT-5.5 Cyber 模型”。企业可以借助该专用模型对自身的基础设施与应用程序开展红蓝对抗，在黑客与恶意攻击者趁虚而入之前提前发现并堵塞系统漏洞。 — Katy Elkin > [05:07](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=307s) 首次公开宣布在欧洲上线“推理数据本地化”，彻底确保支撑各项 AI 计算的物理 GPU 算力均直接部署并留在欧洲境内。 — Katy Elkin > [05:07](https://www.youtube.com/watch?v=ykNmTvRcqPQ&t=307s) OpenAI 的终极愿景是依托企业级智能平台，协助机构全面管理、监控并规模化拓展 AI 价值。 — Katy Elkin ## Mentioned Katy Elkin, OpenAI, NatWest, Cora+, 澳大利亚联邦银行 (Commonwealth Bank of Australia / CBA), ChatGPT Enterprise, Revolut, Rita, GPT-5.5, GDP-Val, Codex Security, 受信任访问计划 (Trusted Access Program), GPT-5.5 Cyber 模型, EKM (企业密钥管理), Inference Residency (推理数据本地化) ## Video description OpenAI’s vision for how financial services firms can use AI across three pillars: transforming workflows, amplifying workforce productivity, and building AI-native customer products. [Source](https://www.youtube.com/watch?v=ykNmTvRcqPQ)

Source quotes

[0:00] Katy Elkin: 首先，如何重塑劳动力，以便让 AI 创造更大的价值？其次，业务工作流该如何重新设计，才能更好地承接并放大 AI 的影响力？最后，如何将 AI 深度注入产品中，从而击败竞争对手并在市场中立于不败之地？ [0:55] Katy Elkin: 风控、合规与日常运营等核心环节长久以来受制于高昂的成本和极高的业务复杂度。如今，AI 使得银行能够从底层彻底重新设计自身的工作流程，实现前所未有的运营方式。 [0:55] Katy Elkin: AI 在金融机构内部正演变为一位“全天候在线的数字同事”。AI Agent 能够自主执行繁琐任务、提炼关键洞察并成倍提升工作产出，从而将企业最核心的专业人才从案头工作中解放出来，专注于维系客户关系和推动营收增长。 [0:55] Katy Elkin: 核心目标在于全面普及全员的“AI 流畅度”，让员工在客户服务、反欺诈、数字银行及中后台运营等日常业务中广泛使用 ChatGPT、应用插件及定制化 Agent。 [2:43] Katy Elkin: 过去只有高净值客户才能享受的高度定制化金融服务，现在可以通过 AI 规模化平铺给所有普通客户。 [2:43] Katy Elkin: 诸如交易领域的 AI 投研辅助、底层代码架构的现代化改造、个性化财富管理等场景，都已不再是空谈，而是正在与客户并肩攻坚的实战项目。 [3:48] Katy Elkin: OpenAI 推出了迄今为止能力最强的模型——GPT-5.5，该模型专为金融分析和专业代码编写等复杂知识工作而设计。 [3:48] Katy Elkin: 在衡量模型完成“具备实际经济价值工作”的基准测试 GDP-Val 中，GPT-5.5 展现出了无可争议的业界领先水平。 [5:07] Katy Elkin: 所有受信任的企业级客户均可申请使用专门的“GPT-5.5 Cyber 模型”。企业可以借助该专用模型对自身的基础设施与应用程序开展红蓝对抗，在黑客与恶意攻击者趁虚而入之前提前发现并堵塞系统漏洞。 [5:07] Katy Elkin: 首次公开宣布在欧洲上线“推理数据本地化”，彻底确保支撑各项 AI 计算的物理 GPU 算力均直接部署并留在欧洲境内。 [5:07] Katy Elkin: OpenAI 的终极愿景是依托企业级智能平台，协助机构全面管理、监控并规模化拓展 AI 价值。

Source quotes

[1:40] Katy Elkin: always-on coworker [1:40] Katy Elkin: AI fluency [3:48] Katy Elkin: downright thrilling

Original transcript

[0:00] [music] [0:02] [applause] [0:05] >> Good afternoon, everyone. Thank you for [0:07] being here with us today. [0:09] It is not often, as Matt said, that we [0:11] get a room of executives like you [0:12] together in person. [0:15] I'm fortunate to work with many of these [0:18] leading companies in our financial [0:19] services team. [0:21] One thing strikes me that comes up from [0:23] these companies over and over again, [0:25] which is they tend to ask the same three [0:27] questions in our conversation. [0:30] First, how can we rethink our workforce [0:33] so that AI can drive more value? [0:36] Second, how can our workflows better [0:38] capture the impact of AI? [0:41] And finally, how can we better infuse AI [0:44] into our products so we can beat our [0:45] competition and win in the markets? [0:48] Let's take a look at how some [0:49] organizations are doing this effectively [0:52] today. [0:55] First, driving value with AI workflows. [0:58] Processes like risk, compliance, and [0:59] operations have long been limited by [1:02] cost and complexity. Now, banks can [1:04] redesign themselves and redesign their [1:06] workflows from the ground up and enable [1:08] themselves to operate in entirely new [1:10] ways. [1:13] NatWest is a great example of this. [1:15] They're partnering with OpenAI to help [1:17] customers tackle everyday financial [1:19] challenges from mortgages to banking to [1:22] money management. [1:24] They're already exploring over 200 AI [1:26] projects [1:27] and over 25 of those are live in [1:29] production today. In fact, Cora Plus, [1:31] their AI assistant, has helped raise [1:34] their customer satisfaction score by [1:36] over 150%, which is pretty remarkable. [1:40] Secondly, multiplying multiplying [1:42] workforce impact. Inside financial [1:45] institutions, AI is becoming an [1:47] always-on coworker, agents that execute [1:49] tasks, surfacing insights, and [1:51] multiplying output. [1:53] So, your best people can spend more time [1:56] building customer relationships and [1:58] focusing on revenue driving activity. [2:02] We see this with Commonwealth Bank of [2:04] Australia. CBA is Australia's largest [2:06] bank and they've rolled out ChatGPT [2:09] Enterprise to 50,000 of their employees. [2:12] Their focus on doing this is AI fluency. [2:15] How can they operate at scale and help [2:17] their team use ChatGPT, apps, and custom [2:20] agents in everyday workflows in order to [2:23] help with customer service, fraud, [2:26] digital banking, and operations. [2:28] They're also bringing a cyber initiative [2:30] to over 1 million small businesses in [2:33] partnership with OpenAI. [2:35] CBA is a strong example of a regulated [2:38] bank rolling out AI firm-wide. [2:43] And third, products. [2:45] AI will redefine how institutions build [2:48] products that serve, advise, and win. [2:51] The level of customization that was once [2:53] only possible to deliver for your [2:55] high-net-worth clients can now be scaled [2:57] out customer-wide. [3:01] A great example of this is Revolut. [3:04] Revolut is working with our team to [3:06] embed AI across the customer experience [3:09] from AI financial AI-powered financial [3:11] crime detection to Rita, their [3:13] AI-powered assistant. [3:15] The result is stronger fraud detection, [3:17] less operational friction, and faster [3:19] support for millions of customers. [3:22] It's pretty amazing. [3:25] And there are many other examples of [3:26] initiatives we're working on with real [3:28] customers both here in Europe and around [3:30] the world. So, things like AI research [3:32] for trading, modernizing code [3:34] infrastructure, anybody here still on a [3:35] mainframe COBOL legacy, [3:37] personalized wealth management. Again, [3:39] these are real initiatives and we're [3:41] tackling these in collaboration with our [3:43] customers. We'd love to work on these [3:44] with any of you. [3:48] Staying on the edge matters more now [3:50] than it ever has before. [3:52] AI is advancing at a faster rate than [3:55] any technology we've seen. [3:57] And there are no signs of it slowing [3:58] down. [4:01] And it all starts with the model. [4:03] GPT 5.5 is our most capable model yet, [4:06] designed for professional work such as [4:08] financial analysis and writing code. [4:12] GDP Val, our evaluation specifically [4:15] designed to test models for their [4:17] ability to complete economically [4:19] valuable work, things like spreadsheets, [4:21] presentations, doing research, clearly [4:25] shows [4:26] that GPT 5.5 is state-of-the-art. Now, I [4:29] started my career as a business analyst, [4:31] and so for me, this is downright [4:33] thrilling. [4:36] We've also worked to make the model more [4:38] intuitive and more natural to interact [4:40] with, while also making it even more [4:42] efficient, so you can get your work done [4:44] fast. [4:45] And while model performance keeps [4:47] improving, we're also driving down the [4:49] cost on every generation. [4:53] Now, as AI models become more [4:55] intelligent [4:56] and capable of advanced coding and [4:58] computer use, [5:00] we know that in financial services, [5:02] leaders like yourselves are more [5:03] concerned about cybersecurity than ever [5:05] before. [5:07] That's why 2 months ago we launched [5:08] Codex Security, a powerful tool to help [5:11] you identify and patch vulnerabilities [5:13] directly in your software. [5:15] And today, we are launching our trusted [5:18] access program to allow any of our [5:20] trusted enterprise customers to access [5:22] our special GPT 5.5 cyber model, so that [5:26] you can red team and catch [5:29] vulnerabilities in your infrastructure [5:30] and applications before malicious actors [5:33] have that opportunity. [5:37] And all of this runs with world-class [5:40] enterprise security and control. [5:42] Your data stays protected with encrypt [5:44] with encryption in transit and at rest, [5:47] as well as support for enterprise key [5:49] management. [5:50] Critical for European institutions, we [5:54] also have extensive support for data [5:56] residency [5:57] with data storage in all of these [5:58] locations. [6:00] And for the first time today, we're [6:02] excited to announce inference residency [6:05] in Europe, ensuring that the GPUs [6:07] powering all of your intelligence [6:09] reside directly in Europe. [6:12] Now, our vision is to help you govern, [6:14] monitor, and scale with AI on an [6:17] enterprise intelligence platform. [6:20] Today is all about sharing that vision [6:22] and showing you how to achieve [6:24] meaningful outcomes across your [6:25] workforce, workflows, and products. [6:29] You'll see it live in the demos, you'll [6:30] hear it directly from customers, and [6:32] you'll even see it directly in how Open [6:34] AI's very own finance team uses our own [6:36] platform and technologies. [6:39] We can't wait to show you what it looks [6:40] like, and we're even more excited to [6:43] build these solutions with you. [6:45] >> [applause]