Japan to Seek Early Warning of Software Flaws From US AI Developers

The measure is expected to be part of an emergency cybersecurity package that the government could unveil at a ministerial meeting as early as Friday, according to Reuters' account of the Nikkei report. Prime Minister Sanae Takaichi told officials in late October to draw up steps to strengthen the country's cyber defences after a rise in attacks.

The request reflects how frontier AI labs have become a major source of vulnerability information. Their most capable models can now search large codebases for exploitable flaws far faster than human researchers. Companies that hold that information before it is disclosed can decide who learns of a bug first, and Tokyo wants Japanese agencies and infrastructure operators to be among them. The reports did not name the companies Japan will approach, how the information would be shared, or whether any developer has agreed.

October showed what late disclosure can cost. Critical vulnerabilities disclosed in PHP, a programming language that underpins much of the web, disrupted government and corporate websites in several countries and prompted millions of hacking attempts, according to Reuters.

The plan builds on other recent steps. Japan has said it will set up a team that uses AI to simulate cyberattacks and find weaknesses in technology systems. According to the Nikkei, it has also asked major US developers, including Anthropic and OpenAI, for wider access to their advanced models in response to cyberattacks believed to have been carried out with AI's help.

Getting access to a model is not the same as receiving its findings. Japan's new request would go further by asking companies to pass on what their systems discover before it is published. How much the US firms are willing to share, and on what terms, should become clearer once the package is released.