It was past two in the morning in Tokyo when the alert landed. A security analyst at one of Japan’s big infrastructure operators stared at a screen full of warnings — another probe against a system that keeps trains running, water flowing, or power humming.
She had seen this before. What she had not seen was a tool that could answer back at the same speed as the attacker. That gap is exactly what Google Fairwind Japan is now trying to close.
On November 16, Google Cloud chief executive Thomas Kurian told Nikkei that Japan’s government and its critical infrastructure operators will get early access to Google’s most advanced artificial intelligence models, tuned for cyber defense. The move brings one of America’s closest allies into a program that, until now, had mostly served the United States.
What Google Fairwind Japan Actually Offers
The offer is an expansion of Google’s Fairwind Program, which launched in September 2026. Fairwind is a vetted-access scheme that gives approved defenders early access to Google’s cyber-specialised AI models, ahead of their release to the general public.
More than 650 partner organisations were already taking part at launch, according to Google. But the access had been extended mainly to United States government agencies and security companies. Japan is now joining that circle, with its National Cybersecurity Office and Digital Agency among the first to benefit, according to Nikkei.
The centrepiece is a cyber-tuned version of Gemini 4 Carbon, Google’s strongest AI model. Google says the model scored 88.2% on SEC-bench Pro, a test of finding real software vulnerabilities, and it is built to hunt for weaknesses in code before an attacker finds them.
Google is also launching a new service in Japan that automatically detects system vulnerabilities, called “Scan for Japan”, according to Nikkei. Together, the model access and the scanning service give Japanese defenders tools they would otherwise have had to wait months to receive.
Access is not open to everyone. Fairwind applicants go through background checks, and Google prioritises governments, critical infrastructure operators, and core technology platforms. The idea is simple: a model powerful enough to find serious flaws is also powerful enough to cause harm in the wrong hands.
Partners may only run dual-use work for defensive purposes, such as authorised threat simulation, reverse engineering, and malware analysis. The guardrails are the price of admission: vetted teams get the frontier model, while everyone else waits for the public release.
Why Japan Is Moving So Fast
The timing is not subtle. Japan logged more cybersecurity incidents in the first three quarters of 2026 than it did in all of 2025, according to Nikkei. The country’s networks are being probed, scanned, and attacked at a pace its defenders have not faced before.
Artificial intelligence is fuelling the surge. AI-assisted attacks have lowered the barrier to entry for cybercriminals, letting less skilled actors do work that once required real expertise. A phishing email that once took hours to craft can now be generated in seconds, tailored to its target.
The threat is not limited to government networks. Operators of power grids, transport systems, telecom networks, and hospitals are among the most consequential targets in any country. Giving them early access to advanced defensive AI suggests Tokyo sees the danger extending well beyond ministries and agencies.
Kurian raised the point himself during his visit to Japan, where he met senior government officials to discuss the country’s national cybersecurity strategy. The message was that speed now matters more than ever: the window for responding to a new threat has shrunk from weeks to hours.
The past week has underlined the point. A recently exploited OpenSSH flaw, under active attack on Linux servers, showed how quickly a single weakness can turn into a global problem. Defenders who find and fix flaws first win; everyone else cleans up afterwards.
The ¥10 Billion Backdrop
Tokyo has already opened its wallet. On November 6, Japan’s cabinet decided to allocate ¥10 billion — about $63 million — from reserve funds in the fiscal 2026 budget to use AI to detect vulnerabilities in key government systems and shore up defenses.
Google’s offer fits neatly into that spending push. The government has funded AI-driven defense, and a major AI provider is now offering early access to the tools that could fill that mandate. It is a rare case of public money and private technology arriving at the same moment.
The relationship predates this week. In 2026, Google partnered with Japan’s National Institute of Information and Communications Technology and the Digital Agency to advance AI-driven detection of software vulnerabilities. The new access framework builds on that groundwork.
All of this sits inside Japan’s broader plan to build an “active cyber defense” framework — a shift from reacting to attacks after they happen toward hunting threats before they land. Early access to frontier models is the sharp end of that plan.
Google Fairwind Japan and the Race Among AI Labs
Japan has not limited its outreach to one company. The government has also sought expanded access to AI from other American firms, including Anthropic and OpenAI, as cyber threats have intensified.
That makes the Fairwind expansion part of a wider contest. OpenAI runs a vetted-defender program of its own, called Daybreak, and Anthropic operates a Cyber Verification Program. Each loosens restrictions or unlocks a cyber-tuned model for approved teams while keeping it away from everyone else.
The competition among US AI firms to become the trusted provider for national cyber defense looks increasingly real. Winning a government’s trust now could mean a long-term role in how that country protects its most sensitive systems.
Google’s own security research arm is also turning up the tempo. On November 13, three days before Kurian’s announcement, Project Zero said it plans to speed up its vulnerability disclosure timelines, moving toward 30 days by 2028. Faster disclosure means faster patching — and less time for attackers to exploit a known flaw.
What Happens Next
Several details are still unclear. Google has not said exactly when Japanese agencies will start using the models, or which infrastructure operators will be first in line. The company also has not said whether other allies will get similar access next.
What is clear is the direction of travel. Governments are no longer waiting for frontier AI to trickle down to them through public releases. They are asking for it early, vetted, and purpose-built for defense — and the biggest AI labs are saying yes.
Regulation is moving in the same direction. New York’s RAISE Act, which takes effect on January 1, 2027, will require the largest AI developers to register with the state and report critical safety incidents within 72 hours. The era of AI operating without government oversight is ending on both sides of the Pacific.
For the analyst on the night shift in Tokyo, the change may be quieter than the headlines suggest. No single model will end cyberattacks. But a defender who can spot a flaw in minutes, rather than days, is a defender with a fighting chance — and that, for Japan, is the whole point.



