• Japan plans to request broader access to advanced AI models from Anthropic and OpenAI to bolster cybersecurity, particularly software-vulnerability detection.
  • The request is part of a cyber package expected next week, building on earlier access granted to Japanese financial institutions.
  • Details on eligible organizations, model versions, safeguards, and funding remain unclear, and the initiative faces U.S. national-security restrictions on foreign access to certain AI models.

Japan is reportedly seeking broader access to advanced artificial intelligence models from U.S. developers Anthropic and OpenAI to strengthen its cybersecurity defenses, according to a Nikkei report cited by Newsquawk. The move, part of a cyber package slated for next week, aims to enhance the detection of software vulnerabilities, according to people familiar with the matter. The request is a policy initiative and not yet a confirmed agreement, and the full government package and developers’ responses remain unconfirmed.

The request builds on earlier access arrangements. In late May, Japanese financial institutions received access to OpenAI’s GPT-5.5 for cyber defense, a step Finance Minister Satsuki Katayama described as a major step toward strengthening financial institutions’ protection against attacks. The forthcoming package is expected to expand that cooperation, but key details—such as which organizations will be eligible, which model versions will be provided, what safeguards will apply, and how the initiative will be funded—are still unknown. The available information does not establish access to model weights or permission for unrestricted deployment.

Strategic Importance

The initiative underscores Japan’s growing focus on using AI for defensive cybersecurity, particularly as cyber threats become more sophisticated. Both Anthropic and OpenAI are major U.S. developers of frontier AI models, and their technologies are increasingly distributed through controlled-access arrangements rather than offered universally. Earlier Japanese access to OpenAI’s models followed Tokyo–Washington discussions, suggesting that broader access may require government-to-government coordination as well as agreements with the companies.

However, the request sits alongside U.S. restrictions that can affect a developer’s ability to supply advanced models abroad. Reuters (TRI) reported this week that the Trump administration ordered Anthropic in June to restrict foreign access to its cyber-capable Mythos 5 and Fable 5 models, citing national-security concerns. How those restrictions apply to the specific access Japan now seeks is not clear from the reviewed sources. The tension between allowing trusted defenders to use powerful AI and preventing the same capabilities from helping attackers is a central policy challenge.

Anthropic, founded in 2021 by former OpenAI researchers, develops the Claude model family and is predominantly enterprise-oriented. Its IPO filing disclosed $16.29 billion in first-half 2026 revenue, versus $1.16 billion a year earlier. OpenAI, led by Sam Altman, reported nearly $50 billion in annualized September revenue and $6.7 billion in second-quarter revenue, according to Reuters. The annualized figure is a snapshot extrapolation, not revenue already earned over a full year. Financial comparisons between the two are complicated by different revenue presentation practices, Reuters cautioned.

What to Watch

The next milestone is Japan’s planned cyber package next week. The decisive details will be who receives access, to which models, under what restrictions, with what funding and on what timetable. Until those terms appear, claims of a completed deal or unrestricted Japanese access would be premature. The most plausible consequence is a wider network of approved Japanese institutions using advanced AI for defensive security, building on the banking precedent. The limiting factors are likely to be developer safeguards, U.S. national-security rules and Japan’s ability to translate model access into practical protection. The reviewed reporting does not provide a quantitative forecast for the initiative’s effect on Japanese GDP, cyber losses, or either developer’s revenue.

Correction: An earlier version of this article misstated the date of the Nikkei report. It was published on October 30, 2026, not October 29.