Data scientists perform last rites for 'dearly departed datasets' in 2nd Trump (apnews.com)

A coalition of data scientists, led in part by the Federation of American Scientists, published a “Dearly Departed Datasets” list documenting federal datasets and tools that have been eliminated, had variables stripped, or had access degraded since President Trump returned to office. Released on Halloween as a pointed provocation, the catalogue groups losses into complete terminations, deleted variables, removed access tools, and datasets that were salvaged and mirrored outside government. Examples include a Census dataset linking income inequality to disaster vulnerability, a health surveillance feed tracking drug-related ER visits, a farm hiring/workhours survey, removal of race and ethnicity fields from a federal workforce dataset, deletion of transgender inmate figures, and excision of gender-identity questions from a crime-victims survey. For the AI/ML community this matters because such changes erode training and evaluation data, reduce feature completeness for fairness and bias audits, and undermine reproducibility of research that relies on public federal sources. Variable deletions and tool removals break longitudinal analyses and automated pipelines; outright terminations force reliance on scraped mirrors with uncertain provenance and licensing. While the list found only dozens of fully terminated datasets among hundreds of thousands produced by federal agencies, authors warn continued risk from staffing losses and shifting priorities — a reminder that model robustness and responsible AI depend on stable, transparent public data infrastructure.