AI News Report — Accurate. Independent. AI-Focused. · https://ainewsreport.org.njangi.app
Lead investigation
Pre-release evaluation of an open-weight model has a property that evaluation of an API-served model does not: once the weights are out, the findings describe something nobody can patch.
AI Safety & Ethics Critical risk 3 min read
Executive policy brief
The obligations on general-purpose model providers have applied since August 2025. The enforcement powers behi...
Safety & ethics spotlight
An agent acting with a user's credentials is, to every access control system in the path, that user. The permi...
The AI literacy obligation in the EU AI Act is short, vague and already applying. Blueprints that treat it as a training procurement exercise are answering a di...
3 min read Global
Open-source projects run on provenance claims. A contributor asserts they wrote the code and hold the rights. Model-assisted contribution makes that assertion h...
3 min read Global
Vendor governance playbooks are useful and interested at the same time. Reading them well means separating the framework from the product it is adjacent to.
3 min read Global
Autonomous agents in the security function are agents with unusually high privilege, operating on the systems that detect misuse. The governance question is who...
3 min read United States Global
Model-level governance asks whether a system should be deployed. Runtime governance asks what it is doing right now. Most organisations have built the first and...
3 min read Global
Counting published frameworks measures disclosure. Whether the frameworks constrain anything is a separate question, and a harder one to answer from outside.
3 min read Global
An aggregate that holds steady can conceal a composition that is changing underneath it. Employment research on AI keeps running into this, and headline numbers...
3 min read United States Global
Inventory, least privilege, egress control, audit. None of these are new ideas. All of them are harder to apply to something that decides what to do next on its...
3 min read Global
A benchmark that has been published long enough to be widely cited has been published long enough to be in the training data.
2 min read Global
A continental strategy is a coordination instrument. It does not build data centres, train regulators or connect anyone to the grid, and reading it as though it...
3 min read African Union
The shortage is not of speakers. It is of digitised text, of evaluation sets that measure the right thing, and of anyone being paid to build either.
3 min read African Union Global
| Act or framework | Jurisdiction | Status | Risk |
|---|---|---|---|
| Texas Responsible Artificial Intelligence Governance Act (TRAIGA) | United States | Enacted | Medium risk |
| Council of Europe Framework Convention on Artificial Intelligence, Human Rights, Democracy and the Rule of Law | European Union, Global | Enacted | Medium risk |
| EU AI Act (Artificial Intelligence Act) | European Union | Enacted | High risk |
| African Union Continental Artificial Intelligence Strategy | African Union | Enacted | Low risk |
| Colorado Artificial Intelligence Act (SB 24-205) | United States | Under Review | Low risk |
