Verified

Goldman Sachs

Scaling
  • Banking
  • 47,500 employees
  • GS AI Platform, built in-house
  • ~40,000 weekly assistant users
  • 12,000+ engineers
  • Evident #12 (2026)

Goldman Sachs deploys AI through the GS AI Platform, an internal gateway built in 2024 that puts OpenAI, Google and Anthropic models behind its firewall, and the GS AI Assistant, used weekly by about 40,000 of its 46,200 employees by Sep 2026. Goldman Sachs gives its 12,000+ engineers coding tools and the Devin agent, is building Claude agents for accounting and client onboarding, and calls OneGS 3.0, launched in Oct 2025, an operating model propelled by AI. Goldman Sachs has published no dollar return from AI.

goldmansachs.com

Their approach to AI

The platform
The GS AI Platform, which Goldman Sachs built and launched in 2024, is the firm's gateway to outside models behind its firewall and carries the GS AI Assistant, translation, data search and agent work.
The models
OpenAI and Google Gemini models are available on the GS AI Platform, Anthropic's Claude was withdrawn for Hong Kong staff in Apr 2026, and Goldman Sachs also runs open-weight models, mostly or entirely U.S.-made (Sep 2026).
The reach
About 40,000 of Goldman Sachs's 46,200 employees used the GS AI Assistant each week, TIME reported in Sep 2026; every employee has had access since Jun 2025, and use passed 1 million prompts a month in Aug 2025.
The catch
Goldman Sachs has published no dollar return from AI, and David Solomon said in Jul 2026 that he could not give a productivity figure he could underwrite and saw no structural change in expenses.

Snapshot

Employees
47,500
Sep 30, 2026; 47,400 at Dec 31, 2025 (2025 Form 10-K)
Technology budget
Not disclosed
Communications and technology expense was $2.17B in 2025 and $1.89B in the first nine months of 2026, up 20%
AI/ML headcount
Not disclosed
12,000+ engineers, about a quarter of staff (Sep 2026); central AI Engineering and Science group under Daniel Marcu
Who owns AI
Marco Argenti, CIO
Management Committee member; co-chairs the Firmwide AI Risk Committee with the chief risk officer
Platform
GS AI Platform, built in-house
Debuted in 2024; the GS AI Assistant runs on it and went firmwide in Jun 2025
Models
OpenAI, Google, Anthropic, open-weight
Plug-and-play by design (Marco Argenti, Mar 2025); Claude withdrawn for Hong Kong staff in Apr 2026
Reach
~40,000 weekly users
Of 46,200 employees (TIME, Sep 2026); 1 million+ prompts a month by Aug 2025
Use cases
100+ specialized versions
Of the assistant on the GS AI Platform (TIME, Sep 2026); six OneGS 3.0 workstreams; no public count of production use cases

Key findings

  1. Goldman Sachs opened the GS AI Assistant to every employee in Jun 2025 after a pilot with about 10,000, use passed 1 million prompts a month in Aug 2025, and TIME reported in Sep 2026 that about 40,000 of 46,200 employees used it each week.

  2. Goldman Sachs's 2025 annual report, published in Mar 2026, calls OneGS 3.0 a new operating model propelled by AI and starts it with six workstreams: client onboarding and KYC, vendor management, regulatory reporting, lending, enterprise risk management and sales enablement, coordinated since May 2026 by a new Office of Transformation.

  3. Goldman Sachs spent about six months to Feb 2026 co-developing Claude agents for trade and transaction accounting and for client vetting and onboarding with embedded Anthropic engineers, and the autonomous coder Devin was broadly available to its engineers by Feb 2026.

  4. Goldman Sachs has put no dollar figure on AI's return: David Solomon said on Jul 14, 2026 that he could not give a productivity figure he could underwrite and that the expense base had not changed structurally, while Marco Argenti told TIME in Sep 2026 that AI coding tools already yield at least 20% efficiency gains for engineers.

  5. Goldman Sachs raised AI oversight to the firmwide level in 2026: its Aug 2026 quarterly filing describes a Firmwide Artificial Intelligence Risk Committee co-chaired by the CIO and the chief risk officer, where its Feb 2026 annual filing had an AI committee reporting to the technology risk committee.

Where Goldman sits

Stage
Scaling

Goldman Sachs sits at Scaling: the GS AI Assistant has been open to every employee since Jun 2025, more than 100 specialized versions run on the GS AI Platform, AI runs in production in coding, translation and data engineering, and CIO Marco Argenti, a Management Committee member, owns the platform and reports its use. Goldman Sachs meets Embedded's weekly-use test, with about 40,000 of 46,200 employees using the assistant each week by Sep 2026 (TIME), but falls short on the other two: the OneGS 3.0 redesign of six workstreams around AI only began in late 2025, and Goldman has reported no financial effect, with David Solomon saying in Jul 2026 that AI had not yet structurally changed the expense base.

How the stages are defined
Ownership
CIO Marco Argenti, on the Management Committee; OneGS 3.0 developed by President John Waldron
Platform
Hybrid: GS AI Platform built in-house; coding tools and agents bought
Models
OpenAI, Google and Anthropic behind one gateway, plus open-weight models
Hosting
Behind the firm's firewall; GenAI Platform team works on AWS; the clouds and data centers serving the models are not disclosed
Reach
About 40,000 weekly users of 46,200 staff (Sep 2026)

How they deploy

Build the gateway, buy the models

Goldman Sachs built the GS AI Platform as a controlled gateway to outside large language models, and Marco Argenti made clear in Jun 2024 that Goldman Sachs would not build its own models from scratch. Argenti told Fortune in Mar 2025 that Goldman Sachs wants to keep models plug-and-play, choosing each by its strengths, such as coding or reasoning, by how easily it can be modified and by its running cost, and avoiding dependence on a single vendor.

Frontier models plus open-weight models

Marco Argenti said in Oct 2026 that frontier models offer the strongest reasoning for unsolved problems, while open-weight models are cheaper for simple tasks and can be retrained on the bank's own knowledge, which helps Goldman Sachs keep control of its intellectual property. Argenti told Axios in Sep 2026 that, with enough safeguards, a bank could run any model regardless of origin, while Goldman's own open-weight models are mostly or entirely U.S.-made.

An operating model propelled by AI

Goldman Sachs launched OneGS 3.0 in an Oct 14, 2025 memo from David Solomon, John Waldron and Denis Coleman, and its 2025 annual report calls it a new operating model propelled by AI. Denis Coleman said in Dec 2025 that each of the six workstreams has a dedicated team that maps pain points and presents a formal investment case, and that Goldman Sachs funds some cases and holds teams accountable for the productivity that follows.

Agents as a digital workforce

Marco Argenti described a hybrid workforce of people and AI agents when Goldman Sachs began piloting Devin in Jul 2025, and called the Claude agents built with Anthropic a digital co-worker for process-heavy roles in Feb 2026. John Waldron told CNBC in May 2026 that he has described Goldman Sachs as a human assembly line and that digital agents will become the bank's robots.

Foundations before scale

Denis Coleman said on Apr 13, 2026 that early OneGS 3.0 learnings pushed Goldman Sachs to double down on the foundations of its infrastructure, accelerating cloud migration and investment in the accuracy, completeness and timeliness of its data, which he called critical to deploying AI across the firm.

Capacity and headcount

Marco Argenti describes the aim of agents as injecting capacity, and said in Oct 2026 that Goldman Sachs may have the option to reduce headcount but that engineering backlogs exceed what each planning cycle funds. Denis Coleman said on Oct 13, 2026 that over time Goldman Sachs should not need the same degree of headcount-related expense to grow.

Deployment model
Goldman Sachs runs AI centrally from its Engineering division under CIO Marco Argenti. Daniel Marcu's AI Engineering and Science group builds the GS AI Platform and AI products; a Jan 2025 memo had Rahul Sharma leading platform engineering and Bing Xiang leading AI research under Marcu, and Archana Vemulapalli has led AI product management and relationships with frontier AI labs since Apr 2026. Engineering teams aligned to Global Banking & Markets, Asset & Wealth Management and The Core bring the tools into each division, and an AI steering group works with risk and control teams to decide which proposals go ahead (Mar 2025). Since May 2026 the OneGS 3.0 workstreams have been coordinated by the Office of Transformation under Chief Administrative Officer Ericka Leslie, who reports to CFO Denis Coleman.
Model providers
Goldman Sachs built the GS AI Assistant on OpenAI, Google Gemini and Meta Llama models (Jan 2025) while evaluating Anthropic, Mistral and Cohere. Anthropic's Claude was later added to the internal platform, and Goldman Sachs withdrew it for Hong Kong staff in Apr 2026, leaving Gemini and ChatGPT available to them. Goldman Sachs engineers use GitHub Copilot (since Jun 2024), Gemini Code Assist (Mar 2025) and Cognition's Devin, and Goldman Sachs co-developed Claude agents with embedded Anthropic engineers for about six months to Feb 2026. David Solomon said on Apr 13, 2026 that Goldman Sachs had Anthropic's Mythos model for cyber defense work, and Marco Argenti said in Sep 2026 that Goldman's open-weight models are mostly or entirely U.S.-made.
Data and infrastructure
Goldman Sachs keeps the GS AI Platform behind its firewall, with a protective layer that insulates the firm's sensitive data from outsiders (Business Insider, May 2025), and Marco Argenti's framework, as described in Sep 2026, keeps models from reaching source data directly, routing them through a bank data platform with access controls. Legend, the data platform Goldman Sachs built over a decade and made public in 2020, now has a natural-language search layer, Legend AI Query. Goldman Sachs has not said which clouds or data centers serve its models: a Goldman Sachs posting for its GenAI Platform team, removed in Aug 2026, asked for serverless AWS architecture skills, and Le Monde Informatique described Argenti's Jan 2024 approach as cloud foundation models paired with smaller open-source models hosted on-premises or in private clouds. Goldman Sachs opened a Bellevue, Washington engineering office for more than 125 AI and cloud employees in Sep 2026.

Platforms and internal tools

  • GS AI Platform2024

    The GS AI Platform is Goldman Sachs's internal gateway to outside large language models, with a security layer that keeps sensitive data from leaving the firm; the GS AI Assistant, Translate AI, Banker Copilot and agent work run on it. A 2026 Goldman Sachs job posting describes the platform as a secure, multi-model backbone for autonomous agents.

    Models
    OpenAI, Google Gemini, Anthropic Claude (withdrawn for Hong Kong staff, Apr 2026)

    Source

  • GS AI AssistantPilot by Jan 2025; firmwide Jun 2025

    The GS AI Assistant is Goldman Sachs's in-house chat assistant for summarizing documents, drafting content, analyzing data and translating, with answers drawn from firm data behind the firewall. Marco Argenti announced the firmwide launch in a Jun 2025 memo and said use passed 1 million prompts a month in Aug 2025, and TIME reported in Sep 2026 that more than 100 specialized versions layer proprietary data and workflows onto the common platform.

    Users
    About 40,000 weekly of 46,200 employees (TIME, Sep 2026); all employees with access since Jun 2025
    Models
    OpenAI, Google Gemini and Meta Llama at launch

    Source

  • Developer copilots and DevinJun 2024 (Copilot); Jul 2025 (Devin pilot)

    Goldman Sachs rolled out GitHub Copilot on the GS AI Platform to more than 9,000 developers, about 80% of its engineers, in Jun 2024, and its coding tools also include Gemini Code Assist (Mar 2025). Goldman Sachs began piloting Cognition's autonomous coder Devin in Jul 2025, planning hundreds of instances under human supervision, and Devin was broadly available to its engineers by Feb 2026.

    Users
    9,000+ developers (Jun 2024); 12,000+ engineers (Sep 2026)
    Models
    GitHub Copilot, Gemini Code Assist, Cognition Devin

    Source

  • Translate AIBy May 2025

    Translate AI is an in-house tool built on the GS AI Assistant that translates text to and from English in seconds or minutes, replacing outsourced translation that could take days; Asset & Wealth Management and Global Investment Research teams use it (May 2025).

    Source

  • Legend AI Query and Legend CopilotBy May 2025

    Legend AI Query lets Goldman Sachs staff search the firm's data in plain English without knowing where it sits in Legend, and was in early rollout in May 2025; Legend Copilot helps data engineers generate data models, APIs and report templates on the Legend platform.

    Source

  • Banker CopilotPilot by May 2025

    Banker Copilot is a Goldman Sachs tool meant to help investment bankers compile client and deal data, analyze filings, draft long documents and summarize notes; a group numbering in the dozens had access in May 2025.

    Source

  • Claude agents for accounting and onboardingIn development (Feb 2026)

    Goldman Sachs and embedded Anthropic engineers spent about six months co-developing autonomous Claude agents for trade and transaction accounting and for client vetting and onboarding; Marco Argenti said in Feb 2026 that they were in the early stages and would launch soon.

    Models
    Anthropic Claude

    Source

Vendors and partners

  • OpenAI2024

    Model provider on the GS AI Platform and in the GS AI Assistant · Goldman Sachs blocked staff access to the public ChatGPT in Feb 2023, then brought OpenAI's GPT models onto its own platform; ChatGPT stayed available to Hong Kong staff after Claude was withdrawn in Apr 2026.

  • Google2024

    Gemini models on the GS AI Platform; Gemini Code Assist for engineers

  • Meta2024

    Llama open-weight models on the GS AI Platform and in the GS AI Assistant at launch

  • AnthropicBy 2025

    Claude on the internal platform, co-developed agents, Mythos for cyber defense, and a joint enterprise AI services company · Anthropic engineers embedded at Goldman Sachs co-developed Claude agents for about six months to Feb 2026; Goldman Sachs is a founding partner, with Blackstone and Hellman & Friedman, of the enterprise AI services company Anthropic announced on May 4, 2026.

  • CognitionJul 2025

    Devin autonomous coding agent · Goldman Sachs does not own a stake in Cognition, according to a person cited by CNBC in Jul 2025.

  • GitHub (Microsoft)Jun 2024

    GitHub Copilot coding assistant

  • Amazon Web Services

    Cloud provider; the GenAI Platform team builds serverless AWS architecture · A Goldman Sachs job posting for the GenAI Platform team, removed in Aug 2026, asked for serverless, cloud-native AWS experience; Goldman Sachs has not said which clouds host its models.

Leadership

AI, data and platform leaders

MA

Marco Argenti

Chief Information Officer

Marco Argenti joined Goldman Sachs as a partner in 2019 from Amazon Web Services, where he had been vice president of technology since 2013, and earlier worked at Nokia. Argenti sits on the Management Committee and the Firmwide Artificial Intelligence Risk Committee, sent the Jun 2025 memo that opened the GS AI Assistant to all staff, and has spoken for Goldman Sachs on Devin, the Anthropic agents and controls on agents.

DM

Daniel Marcu

Partner and Global Head of AI Engineering and Science

Daniel Marcu joined Goldman Sachs's Engineering division in Mar 2025 from Amazon's Artificial General Intelligence organization, where he was vice president of Web and Knowledge Services, to lead the development and deployment of AI platforms and products. Marcu co-founded and ran Language Weaver, was chief science officer at SDL, and spent more than 20 years at USC's Information Sciences Institute.

AV

Archana Vemulapalli

Partner and Global Head of AI Product Management and Strategic Relations

Archana Vemulapalli joined Goldman Sachs in Apr 2026 from AMD, where she led enterprise AI commercialization, and earlier ran product and strategy for data and AI at AWS and was chief technology officer of the District of Columbia. Vemulapalli is based in Dallas, reports to Daniel Marcu, and leads Goldman's engagement with frontier AI labs and AI application providers.

NR

Neema Raphael

Chief Data Officer and Head of Data Engineering

Neema Raphael is Goldman Sachs's chief data officer and head of data engineering (Sep 2025), and Goldman's Aug 2026 filing says the chief data officer co-chairs the Firmwide Data Governance Committee. Raphael described Legend AI Query, the natural-language layer over the firm's Legend data platform, to Business Insider in May 2025.

AG

Alex Golten

Chief Risk Officer

Alex Golten joined Goldman Sachs in 1997 in the Credit Department, was chief credit risk officer and head of market, finance and enterprise risk, and is now chief risk officer. Golten co-chairs the Firmwide Artificial Intelligence Risk Committee with Marco Argenti, and Goldman's independent Model Risk group reports to him.

EK

Evan Kotsovinos

Partner and Head of Asset & Wealth Management Engineering

Evan Kotsovinos joined Goldman Sachs in Jul 2026 from Google, where he was head of privacy, safety and security and led work on AI safety and security, after running technology infrastructure at American Express. Kotsovinos's priorities at Goldman Sachs include accelerating AI development and adoption, modernizing platforms and driving cloud migration in Asset & Wealth Management.

Executives

DS

David Solomon

Chairman and Chief Executive Officer

David Solomon was Goldman Sachs's president and COO from Jan 2017 to Sep 2018 and co-head of investment banking from 2006 to 2016, and joined as a partner in 1999. Solomon said on the Jul 14, 2026 call that he could not give an AI productivity figure he could underwrite, and called on Oct 13, 2026 for rigorous testing, independent review and clear human accountability in AI.

JW

John Waldron

President and Chief Operating Officer

John Waldron is president and COO, sits on the board and the Management Committee, and co-chairs the Firmwide Enterprise Risk Committee. Goldman Sachs's 2026 proxy credits Waldron with leading the development of OneGS 3.0 and overseeing automation and AI efforts across the firm, and he told CNBC in May 2026 that most recent job cuts had little to do with generative AI.

DC

Denis Coleman

Chief Financial Officer

Denis Coleman joined Goldman Sachs in 1996, co-headed the Global Financing Group from 2018 and became CFO after serving as deputy CFO from Sep to Dec 2021. The 2026 proxy says Coleman partners with the COO on OneGS 3.0 and guides the prioritization of engineering investments, and the Office of Transformation reports to him through Ericka Leslie.

EL

Ericka Leslie

Chief Administrative Officer; head of the Office of Transformation

Ericka Leslie was named chief administrative officer in May 2026, reporting to Denis Coleman, with responsibility for a new Office of Transformation that coordinates the six OneGS 3.0 goals and their workstreams. Leslie joined Goldman Sachs in 1996, moved into technology in 2001, and was previously COO of Global Banking & Markets and global head of its Operations and Engineering.

MN

Marc Nachmann

Global Head of Asset & Wealth Management

Marc Nachmann joined Goldman Sachs in 1994, co-headed investment banking and then the Global Markets Division, and now runs Asset & Wealth Management. Nachmann spoke for Goldman Sachs on the May 2026 Anthropic services venture, saying a model alone does not change how a company operates, and on the Jul 2026 launch of the AlphaAI investing platform.

AI org map

Central AI engineering in the Engineering division under the CIO; AI-led process redesign run from the Executive Office

Goldman Sachs runs AI from its Engineering division, led by CIO Marco Argenti, a Management Committee member who joined from Amazon Web Services in 2019. Daniel Marcu, global head of AI Engineering and Science since Mar 2025, leads the teams that build the GS AI Platform and AI products, and Archana Vemulapalli, who joined in Apr 2026, reports to Marcu and leads relationships with frontier AI labs. OneGS 3.0, the AI-driven operating model, was developed by President and COO John Waldron with CFO Denis Coleman, and since May 2026 its goals and workstreams have been coordinated by the Office of Transformation under Chief Administrative Officer Ericka Leslie, who reports to Coleman.

Goldman Sachs employs more than 12,000 engineers, about a quarter of its roughly 47,500 staff (Sep 2026), and has not published a count of AI and machine learning specialists. Goldman Sachs's central AI work sits in Daniel Marcu's AI Engineering and Science group inside the Engineering division, alongside engineering teams aligned to Global Banking & Markets, Asset & Wealth Management and The Core, and a Bellevue, Washington office opened in Sep 2026 for more than 125 employees focused on AI and cloud transformation. Evident's 2026 AI Index ranked Goldman Sachs 19th of 50 banks on AI talent, down from 7th.

Reporting lines

  • Chairman and CEO

    David Solomon

    • President and Chief Operating Officer

      John Waldron

      Led the development of OneGS 3.0 and oversees automation and AI efforts across the firm (2026 proxy)

    • Chief Financial Officer

      Denis Coleman

      Partners with the COO on OneGS 3.0 and guides the prioritization of engineering investments (2026 proxy)

      • Office of Transformation

        Ericka Leslie, Chief Administrative Officer

        Coordinates the six OneGS 3.0 goals and their workstreams (May 2026 memo)

    • Engineering division

      Marco Argenti, Chief Information Officer

      Goldman Sachs employs 12,000+ engineers across the firm; Argenti sits on the Management Committee; the line to the CEO is inferred, not published

      • AI Engineering and Science

        Daniel Marcu

        Develops and deploys AI platforms and products (Mar 2025 press release)

        • GS AI Platform engineering

          Rahul Sharma

          Reporting line per a memo reported by Reuters in Jan 2025

        • AI Research

          Bing Xiang

          Reporting line per a memo reported by Reuters in Jan 2025

        • AI Product Management and Strategic Relations

          Archana Vemulapalli

          Leads engagement with frontier AI labs and AI application providers (Apr 2026)

      • Data Engineering

        Neema Raphael, Chief Data Officer

        Raphael heads data engineering as chief data officer; reporting line not published

    • Risk Division

      Alex Golten, Chief Risk Officer

      The chief risk officer reports to the CEO and to the board's Risk Committee; Golten co-chairs the Firmwide Artificial Intelligence Risk Committee with Argenti

    • Asset & Wealth Management

      Marc Nachmann

      Nachmann sits on the Management Committee; reporting line not published

      • Asset & Wealth Management Engineering

        Evan Kotsovinos

        Works across the division, partnering with Foundation Engineering, on AI adoption, platforms, data and cloud (Jul 2026); reporting line not published

      • AI for Asset Management (AlphaAI)

        Lou D'Ambrosio

        AI investing platform launched Jul 2026; reporting line not published

Goldman Sachs's leadership page, checked Oct 31, 2026, lists Marco Argenti on the Management Committee but not among the executive officers, and the firm has not published to whom Argenti reports. Goldman Sachs's filings say the chief technology officer co-chairs the Firmwide Technology Risk Committee with the CISO, who reports to Argenti, but neither the filings nor the leadership page name the chief technology officer.

Where the AI workforce sits

  • Engineers across Goldman Sachs

    About a quarter of the workforce (Sep 2026). Engineers were the first group at Goldman Sachs to get generative AI tools, and Marco Argenti said in Oct 2026 that working with AI is now standard for them.

    12,000+ engineers

  • AI Engineering and Science (platform, research, AI products)

    Daniel Marcu's central group, with platform engineering, AI research and AI product management teams; Goldman Sachs has not published its size.

    Not disclosed

  • The Core Engineering

    Serves the functional areas that give Goldman Sachs its framework, oversight and foundational processes (Sep 2026 press release); The Stack put it at more than 2,000 engineers in Sep 2026.

    2,000+ engineers

  • Anthropic engineers embedded at Goldman Sachs

    Embedded for about six months by Feb 2026 to co-develop Claude agents for accounting and onboarding.

    Not disclosed

Named people

Executive sponsors
  • David Solomon

    Chairman and CEO

  • John Waldron

    President and COO · Led the development of OneGS 3.0 (2026 proxy)

  • Denis Coleman

    Chief Financial Officer · Partners with the COO on OneGS 3.0; guides engineering investment priorities

Engineering division and AI Engineering and Science
12,000+ engineers across the firm (Sep 2026)
  • Marco Argenti

    Chief Information Officer

  • Daniel Marcu

    Global Head of AI Engineering and Science

  • Archana Vemulapalli

    Global Head of AI Product Management and Strategic Relations · Reports to Daniel Marcu

  • Rahul Sharma

    Head of GS AI Platform engineering · Reporting to Daniel Marcu, per a memo reported by Reuters in Jan 2025

  • Bing Xiang

    Head of AI Research within Engineering · Reporting to Daniel Marcu, per a memo reported by Reuters in Jan 2025

Data
Divisional engineering
  • Melissa Goldman

    Global Head of Engineering, Global Banking & Markets · Title as of 2025

  • Evan Kotsovinos

    Head of Asset & Wealth Management Engineering (since Jul 2026)

OneGS 3.0 transformation
Risk and controls
  • Alex Golten

    Chief Risk Officer · Co-chairs the Firmwide Artificial Intelligence Risk Committee

Asset & Wealth Management

AlphaAI is an investing platform built around the sector shifts AI may cause, not an internal productivity tool.

  • Marc Nachmann

    Global Head of Asset & Wealth Management

  • Lou D'Ambrosio

    Chairman of Artificial Intelligence for Asset Management; leads AlphaAI · Appointed in Jul 2026, per a memo reported by Reuters

Use cases

Goldman Sachs has not published a count of AI use cases in production; TIME reported in Sep 2026 that more than 100 specialized versions of the GS AI Assistant run on its platform, and the public record shows coding tools and the Devin agent, translation and data search, a banker copilot pilot, Claude agents in development for accounting and onboarding, and six OneGS 3.0 workstreams. Goldman Sachs has reported a return for one use case: Marco Argenti said in Sep 2026 that AI coding tools yield at least 20% efficiency gains. Goldman Sachs has given no firm-level dollar figure, and David Solomon said on Jul 14, 2026 that the expense base had not changed structurally.

Use caseAreaStatusROI
GS AI Assistant

The GS AI Assistant lets every Goldman Sachs employee summarize documents, draft content, analyze data and translate using several outside models behind the firewall; use passed 1 million prompts a month in Aug 2025.

FirmwideIn production
Translate AI

Translate AI translates text to and from English in seconds or minutes for Asset & Wealth Management and Global Investment Research teams, work that was previously outsourced and could take days.

Asset & Wealth Management and ResearchIn production
AI coding assistants

Goldman Sachs developers use GitHub Copilot, rolled out to more than 9,000 of them in Jun 2024 on the GS AI Platform, and Gemini Code Assist; generated code goes through the same quality checks as other code.

EngineeringIn productionEngineering efficiency up at least 20% from AI coding tools, per CIO Marco Argenti (Sep 2026)Source
Devin autonomous coding agent

Devin takes on tasks such as moving internal code to newer programming languages under human supervision; Goldman Sachs planned hundreds of instances in Jul 2025 and made Devin broadly available to engineers by Feb 2026.

EngineeringIn production
Legend Copilot

Legend Copilot helps Goldman Sachs data engineers maintain the Legend data platform, generating data models, APIs and report templates.

Engineering and dataIn production
Legend AI Query

Legend AI Query lets Goldman Sachs staff search the firm's data in plain English without knowing where it is stored; Business Insider described it as in early rollout in May 2025.

FirmwidePilot
Trade and transaction accounting agents

Goldman Sachs is co-developing autonomous Claude agents with embedded Anthropic engineers to resolve trade reconciliation and other accounting work faster; Marco Argenti said in Feb 2026 that they were in early stages.

OperationsPilot
Client vetting and onboarding agents

Goldman Sachs is building Claude agents to vet and onboard clients faster, work that falls in the client onboarding and KYC workstream of OneGS 3.0; Marco Argenti said in Feb 2026 that it was premature to expect job losses from the agents.

Operations and compliancePilot
Banker Copilot

Banker Copilot is meant to help investment bankers compile client and deal data, analyze filings, draft long documents and summarize notes; a group numbering in the dozens had access in May 2025.

Global Banking & MarketsPilot
OneGS 3.0 process redesign

Goldman Sachs is redesigning vendor management, regulatory reporting, lending, enterprise risk management and sales enablement around AI, alongside client onboarding and KYC; its 2025 annual report says teams already see opportunities in each area.

Firmwide operationsAnnounced
Agents that monitor markets, transactions and client assets

Goldman Sachs's next step after the assistant is agents that continuously analyze markets, transactions and client assets to inform decisions, TIME reported in Sep 2026.

FirmwideAnnounced
Claude Mythos for cyber defense

Goldman Sachs is testing Anthropic's Mythos model to understand its effect on financial-system defenses, working with Anthropic and its security vendors; David Solomon said on Apr 13, 2026 that the firm had the model.

CybersecurityPilot

Adoption

Goldman Sachs spread AI in stages: engineers first, with GitHub Copilot for more than 9,000 developers, about 80% of its engineers, in Jun 2024; then the GS AI Assistant, which reached about 10,000 employees by Jan 2025 and every employee in Jun 2025; then agents, from the Devin pilot in Jul 2025 to Claude agents for accounting and onboarding in 2026. Use of the GS AI Assistant passed 1 million prompts a month in Aug 2025, and TIME reported in Sep 2026 that about 40,000 of Goldman's 46,200 employees used it each week, asking millions of questions a month. Marco Argenti said in May 2026 that Goldman Sachs can see each employee's AI use but does not track it, watching instead how fast teams move from idea to production, and said in Oct 2026 that working with AI is standard for the firm's 12,000+ developers.

By the numbers

Weekly GS AI Assistant users
~40,000
Of 46,200 employees, TIME, Sep 22, 2026; every employee has had access since Jun 2025
Engineers with AI coding tools
12,000+
Access per TIME (Sep 2026); about a quarter of staff (Goldman Sachs, Sep 2026); AI is standard for them, Marco Argenti said in Oct 2026
Engineering efficiency from AI coding tools
20%+
At least 20%, per CIO Marco Argenti to TIME (Sep 2026); no method published
Specialized versions of the GS AI Assistant
100+
Layering proprietary data and workflows onto the GS AI Platform (TIME, Sep 2026)

Rollout, first pilot to today

  1. Feb 2023

    Goldman Sachs blocked staff access to ChatGPT through an automatic restriction on third-party software, as reported by Bloomberg, while building its own generative AI tools.

    Fortune via Yahoo Finance

  2. Nov 2023

    Goldman Sachs had about a dozen generative AI projects, the most advanced in code generation and documentation, and none client-facing.

    Le Monde Informatique

  3. Jun 2024

    Goldman Sachs rolled out GitHub Copilot on its GS AI Platform to more than 9,000 developers, about 80% of its engineers.

    Fortune

  4. Jan 2025

    Goldman Sachs had released the GS AI Assistant to about 10,000 employees, built on OpenAI, Google Gemini and Meta Llama models, with a goal of reaching all knowledge workers in 2025.

    CNBC via NBC News

  5. Mar 2025

    Daniel Marcu joined Goldman Sachs from Amazon as global head of AI Engineering and Science.

    Goldman Sachs

  6. May 2025

    Business Insider reported that Goldman Sachs had built the GS AI Assistant, Banker Copilot, Translate AI, Legend AI Query and Legend Copilot on its GS AI Platform.

    Business Insider via AOL

  7. Jun 2025

    Marco Argenti announced in a memo that the GS AI Assistant was available firmwide after a pilot with about 10,000 employees.

    Reuters via Yahoo Finance

  8. Jul 2025

    Goldman Sachs began piloting Cognition's autonomous coder Devin alongside its 12,000 human developers, planning hundreds of instances.

    CNBC via NBC Chicago

  9. Aug 2025

    Use of the GS AI Assistant passed 1 million prompts a month, Marco Argenti said.

    Börsen-Zeitung

  10. Oct 2025

    David Solomon, John Waldron and Denis Coleman launched OneGS 3.0, saying Goldman Sachs would constrain headcount growth through year-end with a limited reduction in roles.

    Business Insider via Yahoo Finance

  11. Feb 2026

    Marco Argenti said Goldman Sachs had spent about six months co-developing Claude agents for accounting and onboarding with embedded Anthropic engineers, and Devin was broadly available to engineers.

    CNBC

  12. Apr 2026

    Archana Vemulapalli joined as global head of AI Product Management and Strategic Relations, reporting to Daniel Marcu.

    Goldman Sachs

  13. May 2026

    Goldman Sachs created an Office of Transformation under Chief Administrative Officer Ericka Leslie to coordinate the OneGS 3.0 goals and workstreams.

    Goldman Sachs

  14. Aug 2026

    Goldman Sachs's quarterly filing described a Firmwide Artificial Intelligence Risk Committee co-chaired by the CIO and the chief risk officer.

    June 2026 Form 10-Q

  15. Sep 2026

    TIME reported that about 40,000 of Goldman's 46,200 employees used the GS AI Assistant each week, and Goldman Sachs opened a Bellevue, Washington office for more than 125 AI and cloud employees.

    TIME

  16. Oct 2026

    Marco Argenti said working with AI is standard for Goldman's 12,000+ developers and that the firm is moving into using AI for growth.

    The Next Web

Economics & controls

Spend and ROI

Technology spending

Goldman Sachs does not publish a technology budget. Goldman's communications and technology expense, one line of its cost base, was $2.17 billion in 2025, up 9% from $1.99 billion in 2024, and $1.89 billion in the first nine months of 2026, up 20% from a year earlier. Denis Coleman said on Apr 13, 2026 that Goldman Sachs is accelerating investment in cloud migration and data quality, and David Solomon said on Sep 16, 2026 that the firm had accelerated some technology investments, one reason he expected third-quarter non-compensation expenses to rise by more than $500 million from the second quarter.

What AI itself costs

Goldman Sachs has not disclosed what it spends on AI models, tokens or its contracts with Anthropic, Cognition and GitHub. Goldman Sachs is a founding partner, with Blackstone and Hellman & Friedman, of the enterprise AI services company Anthropic announced on May 4, 2026, which CNBC, citing The Wall Street Journal, put at $1.5 billion of commitments from the participating firms; Goldman's share has not been disclosed.

Reported ROI

Goldman Sachs has not published a dollar return from AI. The one quantified return on the record is Marco Argenti's statement to TIME in Sep 2026 that AI coding tools already produce at least 20% efficiency gains for Goldman's engineers, a lower bound given without a method. David Solomon declined on Jul 14, 2026 to give a productivity figure he could underwrite, and said Goldman's better efficiency ratio reflected higher revenue per person and slower headcount growth, not a structural change in the expense base. Goldman's efficiency ratio was 59.4% for the first nine months of 2026, against 62.1% a year earlier and a target of about 60%, a gain Goldman Sachs has not attributed to AI, and Solomon said in Sep 2026 that management had reviewed OneGS 3.0 KPIs at an offsite without publishing them.

Productivity markers

Melissa Goldman, global head of engineering for Global Banking & Markets, told Business Insider in May 2025 that engineers using the developer copilot saw efficiency gains of up to 20%, and Marco Argenti said in Jul 2025 that Devin could lift productivity by up to three or four times the rate of earlier AI tools. Argenti said in Oct 2026 that some Goldman Sachs teams finished a three-month project in two months and that projects once rejected under zero-based budgeting now pay for themselves, and that he judges AI by how fast teams move from idea to production, not by individual usage (May 2026).

Headcount

Goldman Sachs had 47,500 employees at Sep 30, 2026, down 2% from 48,300 a year earlier; headcount fell 3% in the first half of 2026 from 47,400 at Dec 31, 2025, then rose 3% in the third quarter with campus hires. The Oct 2025 OneGS 3.0 memo said Goldman Sachs would constrain headcount growth through year-end with a limited reduction in roles, and the firm recorded about $250 million of severance in 2025, largely for headcount reductions. John Waldron said in May 2026 that most recent job cuts had little to do with generative AI at that point, while Denis Coleman said on Oct 13, 2026 that over time Goldman Sachs should not need the same degree of headcount-related expense to grow.

Governance and risk

A firmwide AI risk committee

Goldman Sachs's 2025 Form 10-K and first-quarter 2026 filing placed AI oversight in a Firmwide Artificial Intelligence Risk and Controls Committee reporting to the Firmwide Technology Risk Committee, which the CISO and the chief technology officer co-chair. Goldman's Aug 2026 quarterly filing describes a Firmwide Artificial Intelligence Risk Committee, co-chaired by the CIO and the chief risk officer and appointed by the chairs of the Firmwide Enterprise Risk Committee, that oversees both the risks of using AI and firmwide risks from AI developments inside and outside the firm.

Model risk review

Goldman Sachs's Model Risk group sits in the second line of defense, is independent of model developers, owners and users, and reports to the chief risk officer; it independently reviews and validates models, keeps a firmwide model inventory and sets usage controls, under the oversight of the Firmwide Model Risk Control Committee (2025 Form 10-K).

Data boundary and model access

Goldman Sachs blocked staff access to the public ChatGPT in Feb 2023 and runs the GS AI Assistant behind its firewall so sensitive data does not leave the firm (Jun 2025). An AI steering group works with risk and control teams to decide which AI proposals proceed (Mar 2025), and Goldman Sachs withdrew Claude for its Hong Kong staff in Apr 2026 after a strict reading of its Anthropic contract, the Financial Times reported, while Gemini and ChatGPT stayed available.

Controls on agents and models

Marco Argenti told the Evident AI Symposium on Oct 22, 2026 that any agent could misbehave, and that IT leaders should contain agents, set rules for them and gate tools before an agent can complete a transaction. Argenti describes the approach as zero trust and defense in depth: Goldman Sachs assumes a model may already be compromised, limits where agents run and what they reach, reviews reasoning chains and uses other models to challenge outputs (Sep and Oct 2026). Devin instances work under human supervision (Jul 2025).

Board oversight and disclosed risks

Goldman Sachs's 2026 proxy says the board and its committees engaged with management on generative and agentic AI throughout 2025, with the Risk Committee overseeing technology risk, including AI. Goldman's 2025 Form 10-K lists AI risks such as incorrect or biased output, leaks of confidential information, opaque models and intellectual property claims, says they may rise with OneGS 3.0, and cites the EU AI Act among the rules that apply. David Solomon said on Oct 13, 2026 that AI needs rigorous testing, independent review, continuous monitoring and clear human accountability, and wrote in the Financial Times on Oct 22, 2026 that individual decision-makers should answer for what their AI systems do, that firms should stress-test AI routinely and reward employees who flag its weaknesses.

Cyber defense

David Solomon said on Apr 13, 2026 that Goldman Sachs is working closely with Anthropic and its security vendors to use frontier models for cyber defense, that it had Anthropic's Mythos model, and that it is accelerating investment in cyber and infrastructure resilience. Goldman's 2025 Form 10-K says AI may increase the frequency and severity of cyberattacks against the firm and its vendors.

Recent signals

  1. Oct 22, 2026

    David Solomon wrote in the Financial Times that individual decision-makers should answer for what their AI systems do, and that firms should stress-test AI and understand the risks they inherit from model developers.

    PYMNTS

  2. Oct 22, 2026

    Marco Argenti told the Evident AI Symposium that any agent could misbehave and should be contained and gated, and that engineers from an AI lab working alongside a company's own staff is the fastest way to build skills.

    Cybersecurity Dive

  3. Oct 13, 2026

    David Solomon called for rigorous testing and clear human accountability in AI on the third-quarter call, and Denis Coleman said Goldman Sachs should need less headcount-related expense to grow over time.

    Q3 2026 earnings call transcript

  4. Oct 9, 2026

    Marco Argenti said AI is standard for Goldman's 12,000+ developers, that some teams finished a three-month project in two months, and that Goldman Sachs is moving to use AI for growth as well as efficiency.

    The Next Web

  5. Oct 6, 2026

    Evident's 2026 AI Index ranked Goldman Sachs 12th of 50 banks, down from 9th, with talent falling to 19th and innovation rising to 7th.

    Evident

  6. Sep 22, 2026

    TIME reported that about 40,000 of Goldman's 46,200 employees use the GS AI Assistant each week, with more than 100 specialized versions on the platform, and Marco Argenti said AI coding tools yield at least 20% efficiency gains.

    TIME

  7. Sep 8, 2026

    Marco Argenti told Axios that, with enough safeguards, a bank could run any model regardless of origin; Goldman's own open-weight models are mostly or entirely U.S.-made.

    DigitalToday, citing Axios

  8. Aug 3, 2026

    Goldman Sachs's June 2026 Form 10-Q described a Firmwide Artificial Intelligence Risk Committee co-chaired by the CIO and the chief risk officer.

    June 2026 Form 10-Q

  9. Jul 14, 2026

    David Solomon declined to give an AI productivity figure on the second-quarter call and said Goldman's expense base had not changed structurally.

    Q2 2026 earnings call transcript

  10. May 12, 2026

    John Waldron told CNBC that digital agents will become Goldman's robots and that most recent job cuts had little to do with generative AI.

    New York Post via AOL

  11. May 5, 2026

    Goldman Sachs named Ericka Leslie chief administrative officer, reporting to Denis Coleman, to run a new Office of Transformation for OneGS 3.0.

    Goldman Sachs

  12. May 4, 2026

    Anthropic, Blackstone, Hellman & Friedman and Goldman Sachs announced an enterprise AI services company that embeds engineers in mid-sized companies to deploy Claude.

    Anthropic

  13. Apr 28, 2026

    Goldman Sachs withdrew access to Anthropic's Claude for its Hong Kong bankers, leaving Gemini and ChatGPT on its internal platform.

    Reuters via Yahoo Finance

  14. Apr 13, 2026

    David Solomon told analysts that Goldman Sachs had Anthropic's Mythos model and was working with Anthropic and its security vendors to use frontier models for cyber defense, days after regulators warned banks about Mythos.

    GovInfoSecurity

  15. Feb 6, 2026

    Marco Argenti told CNBC that Goldman Sachs is building Claude agents for trade accounting and client onboarding with embedded Anthropic engineers, and that Devin is broadly available to its engineers.

    CNBC

Quick answers

How does Goldman Sachs deploy AI?
Goldman Sachs deploys AI through the GS AI Platform, an internal gateway built in 2024 that puts OpenAI, Google and Anthropic models behind its firewall, and the GS AI Assistant, used weekly by about 40,000 of its 46,200 employees by Sep 2026. Goldman Sachs gives its 12,000+ engineers coding tools and the Devin agent, is building Claude agents for accounting and client onboarding, and calls OneGS 3.0, launched in Oct 2025, an operating model propelled by AI. Goldman Sachs has published no dollar return from AI.
What does Goldman Sachs use AI for?
Goldman Sachs has not published a count of AI use cases in production; TIME reported in Sep 2026 that more than 100 specialized versions of the GS AI Assistant run on its platform, and the public record shows coding tools and the Devin agent, translation and data search, a banker copilot pilot, Claude agents in development for accounting and onboarding, and six OneGS 3.0 workstreams. Goldman Sachs has reported a return for one use case: Marco Argenti said in Sep 2026 that AI coding tools yield at least 20% efficiency gains. Goldman Sachs has given no firm-level dollar figure, and David Solomon said on Jul 14, 2026 that the expense base had not changed structurally.
Who leads AI at Goldman Sachs?
AI at Goldman Sachs is led by Marco Argenti (Chief Information Officer); Daniel Marcu (Partner and Global Head of AI Engineering and Science); Archana Vemulapalli (Partner and Global Head of AI Product Management and Strategic Relations). Ownership of AI on the org chart: CIO Marco Argenti, on the Management Committee; OneGS 3.0 developed by President John Waldron.
What AI platform does Goldman Sachs use?
The GS AI Platform, which Goldman Sachs built and launched in 2024, is the firm's gateway to outside models behind its firewall and carries the GS AI Assistant, translation, data search and agent work.
Which AI models does Goldman Sachs use?
OpenAI and Google Gemini models are available on the GS AI Platform, Anthropic's Claude was withdrawn for Hong Kong staff in Apr 2026, and Goldman Sachs also runs open-weight models, mostly or entirely U.S.-made (Sep 2026).
How many Goldman employees use AI?
About 40,000 of Goldman Sachs's 46,200 employees used the GS AI Assistant each week, TIME reported in Sep 2026; every employee has had access since Jun 2025, and use passed 1 million prompts a month in Aug 2025.
How far has AI spread at Goldman Sachs?
Goldman Sachs sits at Scaling: the GS AI Assistant has been open to every employee since Jun 2025, more than 100 specialized versions run on the GS AI Platform, AI runs in production in coding, translation and data engineering, and CIO Marco Argenti, a Management Committee member, owns the platform and reports its use. Goldman Sachs meets Embedded's weekly-use test, with about 40,000 of 46,200 employees using the assistant each week by Sep 2026 (TIME), but falls short on the other two: the OneGS 3.0 redesign of six workstreams around AI only began in late 2025, and Goldman has reported no financial effect, with David Solomon saying in Jul 2026 that AI had not yet structurally changed the expense base.

Open questions

  • What business return has Goldman Sachs measured for the GS AI Assistant, its coding tools and Devin, given that it has published no dollar or productivity figure for any of them?
  • When will Goldman Sachs publish the OneGS 3.0 KPIs that management reviewed in 2026, and what savings or capacity has each of the six workstreams produced?
  • Have Goldman's Claude agents for trade accounting and client onboarding gone into production, and with what effect on its accounting and compliance staff and outside providers?
  • Which clouds and data centers run the models behind the GS AI Platform, and which open-weight models does Goldman Sachs use for commercial work?
  • What are the more than 100 specialized versions of the GS AI Assistant, which of them run in production, and has Goldman Sachs measured a return for any of them?
  • How does Goldman Sachs measure the at least 20% engineering efficiency gain Marco Argenti cites, and has it changed engineering hiring or spending?
  • How many AI and machine learning specialists does Goldman Sachs employ, and how large is Daniel Marcu's AI Engineering and Science group?
  • What does Goldman Sachs spend on AI models, tokens and its Anthropic, Cognition and GitHub contracts, and what is its share of the Anthropic services venture?
  • Does Claude remain available to Goldman Sachs staff outside Hong Kong after the Apr 2026 contract review?
  • Who is the chief technology officer that Goldman Sachs's filings say co-chairs the Firmwide Technology Risk Committee, and to whom does Marco Argenti report?
  • How many Devin instances run at Goldman Sachs, against the hundreds Marco Argenti planned in Jul 2025?

Sources

Goldman Sachs

News

Data confidence

Confirmed
Stated in a primary or reputable public source.
Estimated
Derived or approximated from public reporting.
Unconfirmed
Reported, but not yet corroborated.

Goldman Sachs profile · Enterprise AI diffusion report by Imagine · facts checked against their sources on Oct 31, 2026

More company profiles

All profiles