Google and Anthropic join Japan's AI transparency code but won't answer individual training-data inquiries

What happened

Google and Anthropic have joined Japan's voluntary AI intellectual property transparency code (the Cabinet Secretariat's Principles-Code for Protection of Intellectual Property and Transparency for the Appropriate Use of Generative AI, published August 2026), but said they cannot answer individual inquiries about training material. That limits a route the code was meant to open for copyright owners seeking evidence for potential infringement claims.

What the code contemplates

Under the code, businesses are encouraged to disclose information about their models, training data, and collection methods, and to respond to rights holders' disclosure requests — including whether a specific URL a requester names was included in training data. The code is non-binding and works on a comply-or-explain basis.

Status and limits

This is a report of the companies' stated positions under a voluntary code, not a court ruling or a finding of infringement. Full MLex article is paywalled; the item rests on MLex's published summary plus Bird & Bird's public explainer of the code's principles.