OpenAI-Hitachi Partnership Caps a Quiet Year of Japan Wins

OpenAI and Hitachi are expanding their partnership to modernize Japan’s mission-critical systems. Hitachi said on June 17, 2026 that joint engineering teams will deploy OpenAI’s Codex AI agent to analyse legacy source code. Hitachi will also gain access to OpenAI’s defensive cyber models through the company’s Trusted Access for Cyber programme.

That announcement is the most visible piece of a quiet, year-long stack OpenAI has built across Japan’s critical infrastructure. NTT Data became OpenAI’s first Japanese distributor for ChatGPT Enterprise in May 2025. SoftBank and OpenAI launched a joint venture called SB OAI Japan in November 2025. And finance minister Satsuki Katayama confirmed on May 29, 2026 that Japan’s three largest banks have access to OpenAI’s GPT-5.5-Cyber model under the same Trusted Access for Cyber programme.

What the Hitachi-OpenAI Expansion Actually Covers

The new work, set out in Hitachi’s June 17 announcement on legacy modernization, runs through three named initiatives. First, joint Forward Deployed Engineer teams from Hitachi and OpenAI will use Codex, OpenAI’s AI agent for software work, to read legacy source code, rebuild high-level specifications from it, and run migration testing for replacement systems. Hitachi’s Modernization Centre of Excellence will package those AI solutions under the brand ‘Modernization powered by Lumada’ and roll them out to customers, starting with financial institutions. The modernization work is one half of the partnership; the other half is cyber defence.

Hitachi will gain access to OpenAI’s cybersecurity AI models through Trusted Access for Cyber, a programme OpenAI introduced on February 5, 2026. The framework sits inside OpenAI’s broader Japan Cyber Action Plan and runs on a defence-focused model called Daybreak, with safeguards, governance, and human oversight written into the deployment. Hitachi’s Cyber Centre of Excellence, an internal cybersecurity expert organisation, will act as Customer Zero and run the validation work on Hitachi’s own systems before the results are packaged for customers.

Insights from both workstreams will feed back into HMAX, Hitachi’s next-generation AI solution suite for social infrastructure, with the company’s Frontier AI Deployment Centre leading the product work. Hitachi’s president and CEO Toshiaki Tokunaga framed the deal as part of a broader AI-transformation push. ‘Modernizing legacy systems and enhancing security are critical management priorities for all enterprises seeking sustainable growth in the age of AI,’ he said in the announcement.

Our work with Hitachi is an important step toward enabling the safer and more practical use of AI in Japan’s critical industries and social infrastructure.

Tadao Nagasaki, president of OpenAI Japan, framed the deal in those terms in the joint announcement. The Hitachi deal builds on a memorandum of understanding the two companies signed in October 2025 to enhance Hitachi’s Lumada data-analytics and IoT platform.

Why Japan Has 15,000 Systems That Need This

The modernization work is sized to a specific problem. Hitachi operates about 15,000 systems that support mission-critical social infrastructure in Japan alone, a figure Tokunaga cited in announcing the deal. Those systems include core banking platforms, energy grid controls, rail operations, and other infrastructure where downtime is not an option. Many run on legacy languages and undocumented code, and the engineers who originally wrote them have retired. That combination is exactly the kind of environment where Codex can earn its keep.

Legacy modernization has historically been slow and expensive, dominated by hand-translated rewrites of COBOL stacks into modern languages. Hitachi’s published technology materials describe a workflow that first reconstructs design documentation from COBOL source code, converts that to Java design documentation, and only then generates the Java code, as set out in Hitachi’s 2026 technology strategy page on physical AI. The new partnership lets Codex take on the reverse-engineering step and the migration testing step, which is where most of the manual time goes. That is the labour-cost compression both companies are after.

Hitachi reported ¥10,586.7 billion in revenue for the fiscal year that ended March 31, 2026, with 606 consolidated subsidiaries and roughly 290,000 employees worldwide. Lumada, the company’s data-analytics and IoT platform, sits at the centre of its digital-services business. The OpenAI work slots into Lumada rather than running alongside it.

The Cyber Layer and the Customer Zero Model

The cybersecurity piece of the deal follows what OpenAI calls the Customer Zero model. Under that pattern, an enterprise tests OpenAI’s defensive AI models inside its own environment first, validates them against real threats, and only then packages the results for external customers. Hitachi’s Cyber Centre of Excellence, an internal team of cyber specialists, will run that validation using OpenAI models delivered through Trusted Access for Cyber. The models will be applied to vulnerability identification, prioritisation, remediation, and validation, with safeguards, governance, and human oversight written into the rollout. Hitachi’s own systems become the proving ground, and customer deployments inherit whatever passes that internal test. That makes the Cyber CoE both an internal customer and a product-development partner for OpenAI’s defensive models.

Trusted Access for Cyber was introduced in an OpenAI blog post on February 5, 2026, and the Hitachi deal is the first major Japanese industrial deployment to use it. The same programme is already giving selected Japanese financial institutions access to OpenAI’s GPT-5.5-Cyber model, which finance minister Satsuki Katayama confirmed on May 29, 2026 after meeting OpenAI chief strategy officer Jason Kwon in Tokyo, per the report on Japanese bank access to GPT-5.5. Japan’s three largest banks, MUFG Bank, Sumitomo Mitsui Banking, and Mizuho Bank, are among the recipients per Nikkei reporting from May 28, 2026.

The Stack OpenAI Has Quietly Built in Japan

The Hitachi expansion is one slice of a much larger pattern. Over the past fourteen months, OpenAI has stitched together a presence across nearly every layer of Japan’s critical infrastructure. Each deal targets a different surface area, but they all funnel into the same OpenAI stack.

• NTT Data partnership (May 1, 2025): NTT Data Group became OpenAI’s first authorised distributor in Japan for ChatGPT Enterprise and rolled out an OpenAI Acceleration Programme to 100 major Japanese corporations.
• Digital Agency deal (October 2025): OpenAI and Japan’s Digital Agency agreed to deploy Gennai, an AI tool powered by OpenAI technology, to government employees for innovative public-sector use cases.
• Hitachi AI data centre partnership (October 21, 2025): Hitachi and OpenAI signed a strategic partnership centred on global AI data centre expansion.
• SB OAI Japan joint venture (November 5, 2025): SoftBank and OpenAI launched a Japan-focused joint venture, set out in the November 2025 SB OAI Japan launch announcement, to market Cristal intelligence, an enterprise AI package, exclusively in Japan from 2026.
• GPT-5.5-Cyber for Japanese banks (May 29, 2026): Finance minister Satsuki Katayama confirmed access for some Japanese financial institutions under OpenAI’s Trusted Access for Cyber programme; per Nikkei reporting from May 28, 2026, MUFG Bank, Sumitomo Mitsui Banking, and Mizuho Bank are among the recipients.
• Hitachi modernization and cyber expansion (June 17, 2026): The Hitachi deal adds legacy-system modernization using Codex and expands cyber-defence work into industrial systems.

The list spans Japan’s financial regulators, three of its largest banks, its central government, its largest telecom operator’s enterprise arm, and now one of its biggest industrial conglomerates. The pattern is unusual because most of OpenAI’s enterprise wins in 2026 have been sold through US-based channel partners and integrators rather than signed directly with national champions, a competitive backdrop detailed in coverage of Anthropic’s pricing pressure on OpenAI’s enterprise plans. In Japan, OpenAI is doing both, selling directly to regulators and ministries while distributing through NTT Data and SoftBank. The hybrid structure is harder to replicate in markets without NTT Data or SoftBank equivalents.

OpenAI executives have publicly framed enterprise as a top priority in 2026, including through CEO comments at industry events in late 2025. The Japan stack covers the same layers, finance, government, industrial infrastructure, and telecom, that any large economy relies on for daily operations. The result is one of the most concentrated national enterprise footprints any AI lab has built.

Hitachi operates in similar social-infrastructure verticals across Europe, North America, and Southeast Asia, and the AI solutions developed through this partnership are not Japan-only by design. SoftBank’s Cristal intelligence product is also pitched at a global enterprise audience rather than a Japanese-only one. That makes the Japan stack both a finished product in its home market and a reference architecture OpenAI can take to other countries where critical infrastructure operators have similar legacy-system problems. Codex’s recent expansion into non-developer roles, with five million weekly active users and non-developers growing more than three times faster than engineers per OpenAI’s own data, as laid out in Codex’s six new business plugins for non-developers, points the same way.

Where HMAX Fits in Hitachi’s Own AI Play

HMAX is Hitachi’s existing brand for its next-generation AI solution suite for social infrastructure. The suite already operates across mobility, energy, and connective industries, with Hitachi Energy launching a dedicated HMAX Energy product in March 2026 for critical energy infrastructure. The OpenAI partnership is designed to make HMAX smarter and broader at the same time.

HMAX is anchored on Lumada 3.0, the current generation of Hitachi’s data-analytics and IoT platform. Hitachi’s published technology strategy frames its focus as physical AI, combining AI with large volumes of field data to deliver reliability and availability in mission-critical applications. That positioning is what lets the company frame HMAX as an industrial AI product rather than a generic chatbot wrapper. The framing also distinguishes HMAX from Codex-style code work and GPT-style chat work.

The OpenAI work is one of three external AI partnerships Hitachi publicly lists as strategic. The other two are with Google Cloud, where work centres on AI agents and operational-technology integration, and with NVIDIA, where Hitachi is building a global AI Factory based on NVIDIA’s reference architecture. The ‘Modernization powered by Lumada’ product family will house the OpenAI engagement’s outputs.

Hitachi is using AI to reform its own operations before selling the same tools externally, in line with its customer-zero approach. The company has built an ‘AI agent factory’ using its GlobalLogic subsidiary to develop internal agents for risk management, quality reviews, and supplier vetting. Those agents are then deployed through an internal ‘AI agent platform’ across the company. The OpenAI partnership extends the same logic into modernization and cyber defence for paying customers.

Hitachi’s Modernization Centre of Excellence will develop AI solutions under the Modernization powered by Lumada brand, with the first wave targeting financial institutions, where legacy core-banking platforms are a known pain point and where Hitachi already has deep customer relationships. The cyber piece runs in parallel, with Hitachi’s Cyber Centre of Excellence running internal tests on TAC-delivered OpenAI models and feeding the validated results into HMAX. After the financial-sector pilot, Hitachi says it will roll the modernization tools out ‘sequentially’ across a wide range of industries. Hitachi’s published technology materials describe the wider rollout as a key pillar of how the company combines AI capabilities with its knowledge of mission-critical systems, from assessment through to strategic operations and maintenance.

What Could Still Slow the Rollout

The deployment runs into three concrete friction points that both companies have flagged in their own materials. The first is the sheer difficulty of reverse-engineering decades of legacy code on live banking and infrastructure systems without disrupting them. Migration testing is meant to catch that, but the testing itself runs on the same live systems the bank or grid operator relies on.

The second is the cyber-defence governance question. OpenAI’s TAC programme ships with stated safeguards, governance, and human oversight, but Japan’s financial regulators are still calibrating how far they will let AI models operate autonomously inside critical systems. A public-private working group was created in Japan this month to deal with cybersecurity dangers to the financial system from a model referred to as Claude Mythos, the same model Anthropic is positioning against OpenAI’s GPT-5.5 family. That kind of regulatory drag could apply to HMAX-bound features as much as to the banks themselves.

The third is competition and pricing pressure. Anthropic has been taking enterprise share from OpenAI, and OpenAI’s own executives have publicly called pricing ‘a huge issue,’ a backdrop that complicates any large new enterprise commitment. Hitachi’s published strategy also names Google Cloud and NVIDIA as core AI partners, which gives the company room to balance workloads across vendors rather than concentrating everything inside OpenAI.

Frequently Asked Questions

What is OpenAI’s Codex doing in the Hitachi deal?

Codex, OpenAI’s AI agent for software work, will analyse the source code of mission-critical legacy systems at Hitachi. The teams will use it to reverse-engineer high-level specifications from existing code and run migration testing for replacement systems, compressing a process Hitachi has historically done by hand.

What is Trusted Access for Cyber?

Trusted Access for Cyber is an OpenAI programme that gives verified defenders controlled access to OpenAI’s defensive AI models. It was introduced on February 5, 2026, and is part of OpenAI’s broader Japan Cyber Action Plan. Hitachi’s Cyber Centre of Excellence will test models delivered through TAC on Hitachi’s own systems first, then package the validated results into HMAX products.

Which Japanese financial institutions are using OpenAI tools?

Finance minister Satsuki Katayama confirmed on May 29, 2026 that some Japanese financial institutions have access to OpenAI’s GPT-5.5-Cyber model. Per Nikkei reporting from May 28, 2026, MUFG Bank, Sumitomo Mitsui Banking, and Mizuho Bank are among the recipients. Katayama declined to name the full list.

What is HMAX and how does OpenAI fit into it?

HMAX by Hitachi is the company’s next-generation AI solution suite for social infrastructure, covering mobility, energy, and connective industries. OpenAI’s modernization and cyber insights will be fed into HMAX through Hitachi’s Frontier AI Deployment Centre, so the OpenAI partnership shows up in HMAX products rather than as a standalone offering.

When does the Hitachi-OpenAI modernization work reach customers?

The joint Forward Deployed Engineer teams began work on the modernization approach at announcement, with the first customer deployments expected to target financial institutions before rolling out to other industries. Hitachi has not published a public timeline for the first commercial deployments.