Who could stop a model release?

The Joint Committee on Human Rights report, adopted September 9 and published September 14, proposes statutory oversight with testing, sanction and remedy powers. Launch blocking or withdrawal would turn on an unacceptable risk to human rights; AISI would advise and the oversight body would decide. It also recommends mandatory access to powerful models for the AI Security Institute. These are recommendations to government, not an enacted law.

The committee's release frames the problem across the supply chain: obligations should follow design and development as well as deployment. It calls for risk-based duties, fewer burdens on low-risk applications, and public consultation over prohibited uses.

The institutional design remains open. The committee allows an existing body to be expanded and AISI functions to be combined with oversight. The useful next record is the government's response, followed by actual bill language specifying who appoints the decision-makers, how decisions are challenged, and what powers survive into law.