OpenAI and Google back shared AI training opt-out signal

The strategic issue here is access to training material, rather than a new model capability. ITmedia reports a common preference mechanism with different implementation schedules, and describes the proposal as a draft rather than an adopted standard. Those distinctions matter: announcing support does not establish universal enforcement or immediate changes to training datasets. ([itmedia.co.jp](https://www.itmedia.co.jp/news/article/2611/01/2000002776/))

Our assessment is that common signals could make data governance more predictable for both publishers and model builders. A shared mechanism would potentially reduce the operational burden of maintaining separate instructions for every laboratory. For competing labs, the important question becomes how reliably their data pipelines preserve those preferences through collection, filtering and subsequent training. That could favor organizations able to document provenance, but this report does not establish such an advantage. Race to AGI readers should therefore treat this as infrastructure for negotiating the use of web content, not evidence that general intelligence has moved closer. The competitive consequences depend on adoption, implementation and the availability of alternative data.