Annex IV of the EU AI Act requires a description of the training, validation and test data used by high-risk AI systems. JDN proposes a five-step method based on tracing the data's lineage.
When an AI makes a mistake with a customer, it is the company that deployed it that pays, not the tool. Europe turns this into a rule on December 9, and JDN lists three questions to ask beforehand.
Google is offering European researchers free compute credits through RAISE, which makes AI easier to access. JDN warns that this creates a dependency that becomes costly when leaving, and that portability should be checked up front.
Anthropic now forbids abusive or cruel behavior toward its models. JDN argues that behind this announcement lies a conceptual shift with very concrete consequences.
The UK's AI minister says 500 new vulnerabilities have been found in government systems. The BBC report gives no further detail beyond the headline.
Several Google DeepMind researchers have left the company in protest at its closer ties with the US military. They are particularly concerned about ensuring real human control over AI systems used in classified military operations.