How does Morgan Stanley govern AI and manage AI risk?
Morgan Stanley's responsible-AI practice is built on testing before deployment and on its existing risk departments, not on a published AI charter. When it put OpenAI's GPT-4 in front of advisors from 2023, it built an evaluation framework that tests every AI use case before release, with advisors and prompt engineers grading summaries and a daily regression suite checking compliance, per OpenAI's December 2024 case study. At board level, the 2026 proxy gives the Operations and Technology Committee oversight of technology strategy 'including trends that may affect such strategy, such as artificial intelligence', and the board held 'deep dives' on AI. Models fall under a Model Risk Management Department inside Risk Management that 'independently validates and approves models for use' and annually recertifies them. For vendors, Madhu Coimbatore, now head of the firmwide AI platform, helped write the FINOS AI governance framework in 2024 and said in June 2026 that the firm uses it in new vendor onboarding. For agents, the design rule is bounded autonomy: the FIXR reconciliation agents keep controllers approving every recommendation, and wealth CIO Hina Shamsi said in October 2026 that agents work within 'clearly defined boundaries'. The 10-K has named generative AI as a source of technology and cyber risk since its 2023 edition. The firm has not published AI principles or an AI-specific audit, and its most recent regulatory record on data is a $35 million SEC penalty in 2022 over customer data on unwiped devices.[1][2][3][4][5][6][7][8][9][10]
Data as of Oct 31, 2026 · Morgan Stanley profile updated Oct 13, 2026 · 16 sources cited on this page · 141 banks profiled on this site · Markdown
Who oversees AI at Morgan Stanley, from the board down?
'Disclosed' is from an SEC filing; 'Stated' is a named executive or a partner's published account.
| Body or role | What it does for AI | Status | Source date |
|---|---|---|---|
| Operations and Technology Committee of the board | Oversees operations and technology strategy, including AI as a trend; reviews the technology budget and significant investments; reviews operational risk, including information technology, cybersecurity and third-party oversight, at least quarterly; met 4 times in 2025 | Disclosed | April 1, 2026[3] |
| Full board | 'Deep dives' on technology initiatives, 'including artificial intelligence, and emerging risks'; cybersecurity reporting covers threats from generative AI | Disclosed | April 1, 2026[3] |
| Model Risk Management Department | A distinct department in Risk Management that sets model risk tolerance, provides effective challenge, independently validates and approves models, annually recertifies and periodically revalidates them | Disclosed | February 19, 2026[4] |
| Firmwide AI platform | Madhu Coimbatore, head of AI development platforms in 2024 and now head of the firmwide AI platform, holds the firm's seat on the FINOS AI governing board | Stated | October 2024; June 2026[5][6] |
| Evaluation framework | Every AI use case is tested before deployment; summarization and translation evals graded by advisors and prompt engineers; daily regression testing for compliant outputs | Stated | December 6, 2024[1] |
| Agent design in wealth | Agents operate within boundaries on what data they can access, how, what actions they take and when a human steps in; Shamsi: 'I can't stress enough the importance of guardrails, permissions, monitoring and human oversight' | Stated | October 26, 2026[8] |
| Agent design in finance | FIXR's agents propose fixes for P&L breaks; controllers review, approve or correct every recommendation, and repeated decisions become fixed rules; 'You still preserve that element of human accountability' | Stated | June 30, 2026[7] |
What has Morgan Stanley said or disclosed about responsible AI, and when?
| Date | Item | What it says |
|---|---|---|
| March 14, 2023 | OpenAI partnership | Wealth management becomes one of a handful of GPT-4 launch organizations and OpenAI's only strategic client in wealth management[2] |
| February 22, 2024 | 2023 10-K | Information security risks 'may be heightened by... reliance on new technologies (such as generative artificial intelligence)'; risk management may not adapt as new technologies such as AI are introduced[9] |
| October 11, 2024 | FINOS AI Readiness Governance Framework | Co-authored with Citi and LSEG; 14 kinds of generative AI threats; the firm already used the draft 'as a set of requirements for data security, privacy, cybersecurity, AI safety and model management' given to employees and vendors[5] |
| December 6, 2024 | OpenAI case study | From 7,000 answerable questions to 'any question from a corpus of 100,000 documents'; more than 98% of advisor teams use the Assistant[1] |
| February 19, 2026 | 2025 10-K | Adds AI and tokenization to laws and regulations the firm must track, and to the technologies risk management may not adapt to[4] |
| June 23, 2026 | FINOS AI Fund | Founding premier member; Coimbatore: 'we actively leverage the FINOS AI Governance Framework in new vendor onboarding'[6] |
What is Morgan Stanley's regulatory record on data, and how does it bear on AI?
On September 20, 2022 the SEC fined Morgan Stanley Smith Barney $35 million for failing, over five years, to protect personal information of about 15 million customers: a moving company with no data-destruction expertise decommissioned thousands of hard drives and servers, some later resold online with unencrypted data. The 10-K still cites human error that 'has led to the loss of the Firm's physical data-bearing devices in the past'.[10][4]
The firm's generative AI concerns, as Coimbatore described them in 2024, start with the same issue: data security can be breached 'through a prompt injection attack, or the loss of data from a vector database or a proprietary model', which he said calls for defense in depth, encryption and role-based access. No enforcement action against Morgan Stanley has concerned AI.[5]
What does the Morgan Stanley profile record on this topic?
The rows below are the entries of the Morgan Stanley AI strategy profile that bear on this page, as of Oct 13, 2026; the profile holds the full record of 32 dated moves, 14 use cases and 37 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| Compliance task automation | Bots handling routine non-financial-risk work. | In production | [11] |
| Mythos beta | Testing Anthropic's vulnerability-finding model. | Pilot | [12] |
| FIXR agentic P&L reconciliation | Three cooperating agents reconcile end-of-day P&L across finance, risk, operations and trade-capture systems; about 1,500 controller hours saved a week. | In production | [7] |
| FINOS AI Governance Framework in vendor onboarding | Used for new AI vendor onboarding; the firm is a founding premier member of the FINOS AI Fund. | In production | [6] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Jun 30, 2026 | FIXR: agentic P&L reconciliation cuts the job in half | Three cooperating agents interpret past guidance, learn from controller behaviour and turn repeated patterns into fixed rules; up to six hours per book becomes two to three, about 1,500 hours a week across roughly 100 controllers. | [7] |
| Jun 23, 2026 | Founding premier member of the FINOS AI Fund | With DTCC, RBC and NatWest; Coimbatore says the firm uses the FINOS AI Governance Framework in new vendor onboarding. | [6] |
| Apr 15, 2026 | Pick: AI is a 'productivity phenomenon'; Mythos in beta | On the Q1 call the CEO says the firm is using a beta of Anthropic's Mythos model. | [12] |
| Jan 15, 2026 | Q4 2025 call: growth plans 'embed' AI; LeadIQ on the revenue side | LeadIQ uses AI to introduce advisors to interested clients; in operations 'we now have one human team and one AI team'. | [13] |
| Aug 15, 2025 | Calm, the firm's architecture-as-code tool, open-sourced through FINOS | Already underpinning well over 2,000 application deployments across firms and more than 1,400 internally; Morgan Stanley, BMO, Citi and RBC joined a June 2025 FINOS effort on vendor-neutral AI adoption standards. | [14] |
| Jul 15, 2025 | Patent granted on LLM conversion of legacy code | US 12,360,791, assigned to Morgan Stanley Services Group: converting a legacy program into a human-language description that can be validated and then rewritten; priority May 2024. | [15] |
| May 20, 2025 | Bots do the 'boring compliance work' | Task automation in non-financial risk; 'capacity creation' as the operating doctrine. | [11] |
| Mar 14, 2024 | Jeff McMillan appointed the firm's first head of firmwide AI | A newly created role 'to ensure the appropriate AI strategy and governance are in place'; McMillan had been chief data and analytics officer; five generative-AI use-case buckets: search, summarisation, interpretation, generating solutions and translation across 53 languages. | [16] |
Does Morgan Stanley have responsible AI principles?
It has not published a standalone AI principles document. Its stated practice is evaluation before deployment (every use case tested, with daily regression suites), model validation by its Model Risk Management Department, board oversight through the Operations and Technology Committee, the FINOS AI governance framework for vendors, and agents built with humans approving actions. Wealth CIO Hina Shamsi summarized it in October 2026 as 'guardrails, permissions, monitoring and human oversight'.
How does Morgan Stanley test AI before using it?
Through evals. OpenAI's December 2024 case study describes Morgan Stanley testing every AI use case before deployment: summarization evals graded by advisors and prompt engineers, translation evals for multilingual clients, evaluation datasets of meeting types for the Debrief note-taker, and a daily regression suite of sample questions to catch non-compliant outputs.
Who oversees AI on Morgan Stanley's board?
The Operations and Technology Committee, chaired by Jami Miscik, oversees the firm's operations and technology strategy including AI, reviews the technology budget and reviews operational risks such as cybersecurity and third-party oversight at least quarterly; it met four times in 2025. The full board also held deep dives on technology initiatives including AI.
How does Morgan Stanley control AI vendors?
By applying the FINOS AI governance framework, which its head of firmwide AI platform Madhu Coimbatore helped write in 2024, to new vendor onboarding. In 2024 he said the draft was already used as requirements for data security, privacy, cybersecurity, AI safety and model management given to employees and vendors, and that many controls must be built in by technology providers.
Are Morgan Stanley's AI agents autonomous?
Deliberately not fully. The FIXR agents that reconcile trading P&L propose fixes that controllers review, approve or correct, and only repeated decisions become automatic rules; the work fell from up to six hours per book to two or three. In wealth management, agents that move money or change beneficiaries for service associates run within defined boundaries on data access, actions and when a human must step in.
- Morgan Stanley uses AI evals to shape the future of financial services
- Key Milestone in Innovation Journey with OpenAI
- Morgan Stanley 2026 Proxy Statement (DEF 14A)
- Morgan Stanley Annual Report on Form 10-K for 2025
- Banker-led group shares advice on gen AI's hallucinations, other risks (FINOS AI Readiness Governance Framework)
- FINOS Launches AI Fund to Amplify the Collective Voice of the Financial Services Industry and Accelerate Responsible Agentic AI Adoption
- Morgan Stanley cut its riskiest reconciliation job in half by making its agents less autonomous
- A look inside the AI playbook at Morgan Stanley's wealth division (Hina Shamsi, CIO of wealth management technology)
- Morgan Stanley Annual Report on Form 10-K for 2023
- Morgan Stanley Smith Barney to Pay $35 Million for Extensive Failures to Safeguard Personal Information of Millions of Customers
- Why Morgan Stanley lets bots do the boring compliance work
- 'AI is our friend': Bank CEOs weigh the tech's risks, rewards
- Morgan Stanley (MS) Q4 2025 Earnings Transcript
- Why Morgan Stanley open-sourced its app development tool
- US 12,360,791 B1: Generating new software code from legacy software code using large language models (Morgan Stanley Services Group)
- 5 questions with ... Morgan Stanley Head of AI Jeff McMillan
When Morgan Stanley moves on AI, you'll read it here first.
when one of the banks we cover moves on AI, it is in the next morning's brief · six sourced stories · 7 am PT · free
plus every regulation, bank and agent page update, the morning after · leave any morning