How does MUFG govern AI responsibly?
MUFG governs AI under the MUFG AI Policy announced on March 18, 2025, which sets nine principles: human-centric, reliability and safety, fairness, privacy, intellectual property, information security, transparency, accountability and AI literacy. Its 2026 annual report on Form 20-F says MUFG has a controls framework for appropriate, reliable and secure AI use but that it may not fully mitigate risks such as biased outputs, data leakage and growing reliance on a limited number of third-party vendors, clouds or models. In practice, MUFG Bank made e-learning mandatory before staff received ChatGPT Enterprise accounts, reaching 100% participation, anonymizes personal and confidential data with Private AI before unstructured data reaches its OCEAN platform, and describes its credit and relationship-manager agents as supporting people who keep the decision. Its home supervisor sets the frame: the FSA's AI Discussion Paper (Japanese March 4, 2025; English April 10) warns of a 'risk of inaction' while flagging generative-AI risks, and on May 22, 2026 the FSA and the Bank of Japan asked financial institutions to take short-term measures, under top management, against threats posed by frontier AI. The Japan Times reports MUFG Bank among the megabanks given access to Claude Mythos for vulnerability finding.[1][2][3][4][5][6][7][8]
Data as of Oct 16, 2026 · MUFG profile updated Oct 13, 2026 · 18 sources cited on this page · 140 banks profiled on this site · Markdown
What policies and controls govern AI at MUFG?
| Item | What it says or does | Date |
|---|---|---|
| MUFG AI Policy | Nine principles: human-centric, reliability and safety, fairness, privacy, intellectual property, information security, transparency, accountability, AI literacy | March 18, 2025[1] |
| 20-F AI risk factor | A controls framework for appropriate, reliable and secure AI may not fully mitigate risks including biased outputs, data leakage and reliance on a limited number of vendors, clouds or models; AI-related rules may evolve differently across jurisdictions | July 6, 2026[2] |
| Mandatory AI training | E-learning required before ChatGPT Enterprise access; 100% participation; AI champions in each department | July 7, 2026[3] |
| Data anonymization | Private AI detects and anonymizes personal and confidential data in emails, call logs, PDFs and chats, on premises, before it feeds generative AI on the OCEAN platform | January 7, 2026[4] |
| Human decision-makers in credit | AI credit experts help draft approval documents; a pilot used more than 1,400 evaluations and comments from experienced staff | August 24, 2026[9] |
| Grounded agents | AI Flowchart grounds the model in an ontology and knowledge graph so it does not invent steps; experts still make standardization decisions (according to AWS) | October 8, 2026[10] |
| Cybersecurity line of command | The Group CIO runs the cybersecurity risk management program; the Group CISO reports to the Group CIO and to the Group CRO on IT security risk | July 6, 2026[2] |
What does Japan's Financial Services Agency expect of MUFG on AI?
| Item | What it says | Date |
|---|---|---|
| FSA AI Discussion Paper | 'Preliminary Discussion Points for Promoting the Sound Utilization of AI in the Financial Sector': encourages AI use, citing the 'risk of inaction', while noting generative-AI risks such as misuse for crime and misinformation | March 4, 2025 (English April 10, 2025)[6] |
| Frontier-AI request | FSA and Bank of Japan ask financial institutions to take 'Short-Term Measures ... in Response to Changes in Threat Posed by Frontier AI' | May 22, 2026 (English June 15)[7] |
| Top-management involvement | The request urged countermeasures 'under the direct involvement of top management' and prioritization if frontier AI uncovers many vulnerabilities | June 3, 2026 (as reported)[8] |
| Minister on Mythos access | Japan given access to Claude Mythos Preview; participants in Glasswing were 'mainly the three megabanks', which described the threat as imminent | June 2, 2026[11] |
What has MUFG not disclosed about AI governance?
MUFG has not published how the AI Policy is enforced, which committee approves high-risk AI uses, or any testing or incident results; the 20-F describes the controls framework only in general terms.[1][2]
On frontier models, MUFG declined to comment on reported Mythos access, so how it uses Mythos or a restricted OpenAI model for cyber defense is not on the record.[12][13]
What does the MUFG profile record on this topic?
The rows below are the entries of the MUFG AI strategy profile that bear on this page, as of Oct 13, 2026; the profile holds the full record of 30 dated moves, 16 use cases and 50 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| MUFG AI Policy and AI risk management | A nine-principle AI Policy and, per the 20-F, a controls framework for appropriate, reliable and secure AI use that the bank says may not fully mitigate risks. | In production | [1][2] |
| Equity and partnership stakes in AI companies | Investments in Sakana AI (twice), LayerX and Noetra, plus partnerships with OpenAI and Google, tie the group's AI roadmap to a handful of outside providers, a concentration the 20-F flags as a risk. | In production | [14][15][16][2] |
| PII anonymisation before unstructured data reaches generative AI | Private AI software detects and anonymises personal and confidential data in emails, call-centre logs, PDFs and chat logs, on premises, before the data is linked to the OCEAN big data platform used for generative and operational AI. | In production | [4] |
| Frontier AI models for vulnerability finding and cyber defence | Nikkei Asia and Reuters report that MUFG Bank, SMBC and Mizuho are expected to gain access to Anthropic's Claude Mythos and a restricted OpenAI model; MUFG declined to comment, and the 20-F separately warns of AI-enabled cyber threats. | Announced | [17][18][13][2] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Jul 6, 2026 | Annual report on Form 20-F: AI risk factor | The 20-F for the year ended 31 March 2026 reports total assets of ¥425.58 trillion, describes the MUFG AI Policy, notes that outsourcing expenses rose 22.0% partly on AI, and adds a risk factor on AI use by MUFG and others. | [2] |
| May 13, 2026 | Reported access to Claude Mythos and an OpenAI cyber-defence model | Nikkei Asia reports MUFG, Mizuho and SMFG are set to gain access to Anthropic's Claude Mythos from as soon as the end of May; Reuters, citing a source, says about two weeks, and that MUFG declined to comment. On 28 May Reuters, relaying Nikkei, reports the three banks are also expected to get a restricted OpenAI model for defence against cyberattacks. | [17][18][13] |
| Jan 7, 2026 | Private AI anonymisation for the OCEAN data platform | MUFG Bank adopts Private AI software to detect and anonymise personal and confidential information in emails, call-centre logs, PDFs and chat logs before they enter its OCEAN big data platform, so generative and operational AI can use unstructured data; testing on simulated business files preceded adoption. | [4] |
| Mar 18, 2025 | MUFG AI Policy announced | A nine-principle policy: human-centric, reliability and safety, fairness, privacy, intellectual property, information security, transparency, accountability and AI literacy. | [1] |
Does MUFG have an AI policy?
Yes. The MUFG AI Policy, announced on March 18, 2025, sets nine principles: human-centric, reliability and safety, fairness, privacy, intellectual property, information security, transparency, accountability and AI literacy. The 2026 Form 20-F adds that MUFG has a controls framework for appropriate, reliable and secure AI use that may not fully mitigate the risks.
What AI risks does MUFG report to the SEC?
Its annual report on Form 20-F, filed July 6, 2026, says AI controls may not fully mitigate risks such as biased outputs and data leakage, that it relies increasingly on a limited number of third-party vendors, clouds or models, that AI-enabled cyber threats are rising, and that AI rules may evolve differently across jurisdictions.
How does MUFG protect customer data when using generative AI?
MUFG Bank uses Private AI software, on premises, to detect and anonymize personal and confidential information in emails, call-center logs, PDFs and chats before that data reaches the OCEAN big data platform used for generative and operational AI.
What has Japan's FSA asked banks to do about frontier AI?
On May 22, 2026 the FSA and the Bank of Japan asked financial institutions to take short-term measures against threats posed by frontier AI, with top management directly involved and vulnerabilities prioritized. The finance minister said on June 2 that Japan had been given access to Claude Mythos Preview and that the three megabanks were the main participants.
Are humans kept in the loop in MUFG's AI?
That is how MUFG describes it. Its AI credit experts help draft credit approval documents for staff who decide, the relationship-manager agents described at the 2026 Investors Day support bankers before, during and after meetings, and AWS says experts still make the decisions in the AI Flowchart standardization tool.
- Announcement of the MUFG AI Policy
- Mitsubishi UFJ Financial Group, Inc. Annual Report on Form 20-F for the fiscal year ended March 31, 2026
- MUFG aims to become AI-native with OpenAI
- MUFJ Bank Adopts Private AI
- MUFG Investors Day 2026: Speech Script
- Publication of AI Discussion Paper: Preliminary Discussion Points for Promoting the Sound Utilization of AI in the Financial Sector
- Request Regarding 'Short-Term Measures for Financial Institutions in Response to Changes in Threat Posed by Frontier AI'
- Japanese government and banks granted access to Claude Mythos
- MUFG Report 2026 (Integrated Report)
- How MUFG Bank aims to cut flowchart work by up to 90% with ontology-grounded AI on Amazon Neptune and Amazon Bedrock
- Press Conference by KATAYAMA Satsuki, Minister of Finance and Minister of State for Financial Services (June 2, 2026)
- Japan megabanks set to win Mythos access after Bessent visit
- Japan's major banks to use OpenAI's new model to thwart cyberattacks, Nikkei reports
- MUFG To Make Additional Investment in Sakana AI
- MUFG Invests in LayerX
- MUFG to Invest in Noetra
- Japan megabanks to gain access to Anthropic's powerful AI model Mythos
- Japan megabanks to gain access to Anthropic's Mythos in about two weeks, source says
When MUFG 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 ET · free
plus every tracker, bank and agent page update, the morning after · leave any morning