Overview
- Google has published a policy paper defending use of publicly available web content for model training as a transformative fair use while saying developers should offer website owners simple machine-readable controls to block training.
- The company proposes a Google‑Extended robots.txt tag and says it is testing an opt-out toggle for publishers as a practical way to give sites choice over whether their content is used.
- Google says it is piloting 'value-exchange' partnerships and has paid for access to specialised non-public content to improve the factuality of grounded AI responses, though it has not released program terms or timelines.
- Regulators and publishers continue to press for stronger safeguards: the UK regulator has ordered opt-outs and attribution and publishers and trade groups are pushing for permission-first access, compensation and legal remedies.
- The dispute could change where publishers get referral traffic and how much training large models costs, and Google’s separate proposal for a federally overseen frontier AI regulator would add a national compliance layer if adopted.