SBI is rolling Claude out across its entire financial group and bolting it onto the on-chain finance rails it has spent years building. Unlike NEC or Fujitsu, this is a user, not a vendor. We compare how Japan, the US and the EU are wiring AI into finance, regulation included.
Japan's government and three megabanks gained access to a frontier AI model for cyber defense, with the digital minister calling to expand it to critical infrastructure. Set against the new US executive order and the EU AI Act, here is how Tokyo's approach differs.
On May 27, Fujitsu announced partnerships with two US AI labs at once, Anthropic and OpenAI. We unpack what this dual strategy means versus NEC's single bet, the Forward Deployed Engineer model, and how Japanese IT giants differ from Microsoft-OpenAI and SoftBank.
OpenAI is moving to offer its newest cyber-focused AI model, GPT-5.5 Cyber, to the Japanese government and industry. Former NSA Director Paul Nakasone, now an OpenAI board member, flew to Tokyo to pitch critical-infrastructure cooperation. We unpack the contrast with Anthropic's tightly controlled Claude Mythos, the China-threat framing, and what it means that OpenAI wants Japan as its first international partner.
Anthropic's Head of Global Affairs Michael Sellitto arrived in Tokyo May 15 and confirmed the company is weighing participation in the cyber defense corporate alliance proposed by Japan's ruling LDP. Modeled on US Project Glasswing, the framework will test Japan's distinctive 'middle path': an AI Promotion Law with no penalties, sitting between the EU AI Act's hard regulation and the deregulatory turn under the Trump administration.
On May 1, Japan's METI met infrastructure operators from the power and gas sectors, and Economy Minister Akazawa asked electricity companies to run emergency system inspections and report within a month. The order covers 24 critical-infrastructure operators in the power sector. The trigger was Anthropic's Claude Mythos Preview, announced April 7, which surfaced a 27-year-old bug chain in OpenBSD's TCP SACK code among others. The UK's AISI reached a more measured verdict. How Japan's penalty-free approach compares with CISA's OT principles and the EU's NIS2.
NEC has begun a strategic partnership with Anthropic, becoming the first Japanese company named as a global partner. The deal embeds Claude Opus 4.7 and Claude Cowork into NEC's BluStellar DX platform for banks, manufacturers, and local governments, and deploys Claude across roughly 30,000 employees. We break down NEC's three-layered AI strategy spanning its own cotomi LLM, the national Physical AI initiative, and now Anthropic, and examine why Tokyo's IT giants are hedging toward Claude over ChatGPT.