MUFG, SMBC, and Mizuho with Claude Mythos: The Silent Automation That Begins in Tokyo and Reaches Chennai

The Japanese Government Confirms Access to Claude Mythos

On June 3, 2026, Japanese Finance Minister Satsuki Katayama confirmed that MUFG Bank, Sumitomo Mitsui Banking Corporation (SMBC), and Mizuho Bank have been granted access to Claude Mythos, Anthropic's cybersecurity model capable of autonomously identifying thousands of vulnerabilities in source code. On the same day, Anthropic announced the expansion of the Project Glasswing to an additional 150 organizations across more than 15 countries, most of which are companies whose systems affect over 100 million people. This access was facilitated by U.S. Treasury Secretary Scott Bessent's visit to Tokyo in May 2026.

The three megabanks had already gained access at the end of May to OpenAI's GPT-5.5-Cyber, a model specialized in threat log analysis, security report summarization, and automated patch generation. This combination creates what Nikkei Asia described as "two independent AI perspectives" on the same systems, positioning the three institutions among the few global financial players with direct access to state-level cybersecurity models.

From Security to the Office: What Mythos Does Beyond Cybersecurity

Once integrated into a bank's infrastructure, Mythos does not differentiate between security log analysis and account opening form analysis. The model applies the same capabilities for textual understanding and code automation to any document within the client's systems, creating functional overlap with the office tasks that Mizuho plans to eliminate.

In February 2026, Mizuho revealed, and Bloomberg confirmed, its intention to cut the equivalent of up to 5,000 administrative jobs in Japan over ten years, reducing a staff of 15,000 back-office employees by up to one-third. The projected investment is between $320 million and $640 million in AI over three years. The targeted processes include document verification for account openings, fund transfers, customer data entry, and manual form review, which are precisely the categories of tasks that models like Mythos and GPT-5.5-Cyber perform as secondary functions to their declared security purpose.

MUFG has gone further: in January 2026, Nikkei Asia reported that the bank began positioning AI assistants in the same way it positions human employees, assigning them tasks such as speechwriting and training new hires.

Emi's Argument and What It Fails to Capture

Morito Emi, responsible for digital strategy at MUFG, argues that "the most powerful combination is not human or AI, but human and AI learning side by side." For Emi, language models are complements to productivity, not replacements. This view is shared by SMBC, which developed an AI assistant modeled after the president and CEO of the group, Toru Nakashima, for internal use by employees in everyday decisions.

Emi's argument is consistent with Japanese legal restrictions: mass layoffs are extraordinarily difficult in Japan, and none of the three megabanks foresee direct headcount cuts. The issue is that Emi identifies the risk on the wrong axis. In Japanese megabanks, the relevant question is not whether there will be mass layoffs, but whether vacant positions will be filled. Mizuho, in announcing its elimination of 5,000 positions, was explicit: the mechanism is a combination of hiring freezes, retirements, and voluntary departures.

Japan has the lowest unemployment rate in the G7 and a structurally tight labor market: companies seldom engage in mass layoffs even during recessions. The preferred adjustment mechanism is the curtailment of new hiring. In a market where the generation of younger banking workers is smaller than that retiring, automation does not compete with layoffs; it competes with hiring. The reduction of 5,000 jobs at Mizuho will occur without any headlines in Tokyo.

India: Where the Impact Arrives Before the Announcements

MUFG, SMBC, and Mizuho operate processing centers in Chennai, Bengaluru, and Hyderabad, where offshore teams execute exactly the back-office tasks that the three banks plan to automate. The introduction of models like Mythos into the systems of the parent companies in Tokyo reduces the demand for manual processing across the chain, including the Indian operations, without any announcement regarding restructuring of units abroad.

Unlike American banks, which have announced explicit cuts with public numbers, Japanese megabanks rarely publish granular headcount data by country. Mizuho was an exception in detailing its clerical automation plan in February. For any IT, consulting, or BPO provider with operations directed at these three institutions in Mumbai or Chennai, the relative transparency from Tokyo carries a clear message: the positions that exist today in these cities are not guaranteed to exist in 2030.

Access to Claude Mythos at the three largest banks in Japan is not a mere security tool upgrade; it is the entry of state-level AI infrastructure within organizations that buy, outsource, and automate work on a global scale. The question any vendor for these three institutions needs to answer is: what will these banks start doing internally that you currently do for them?