When AI Benchmarks Plateau: A Systematic Study of Benchmark Saturation

Artificial Intelligence (AI) benchmarks play a central role in measuring progress in model development and guiding deployment decisions. However, many benchmarks quickly become saturated, meaning that they can no longer differentiate between the best-performing models, diminishing their long-term value. In this study, we analyze benchmark saturation across 60 Large Language Model (LLM) benchmarks selected from technical reports by major model developers. To identify factors driving saturation, we characterize benchmarks along 14 properties spanning task design, data construction, and evaluation format. We test five hypotheses examining how each property contributes to saturation rates. Our analysis reveals that nearly half of the benchmarks exhibit saturation, with rates increasing as benchmarks age. Notably, hiding test data (i.e., public vs. private) shows no protective effect, while expert-curated benchmarks resist saturation better than crowdsourced ones. Our findings highlight which design choices extend benchmark longevity and inform strategies for more durable evaluation.

Who Evaluates AI's Social Impacts? Mapping Coverage and Gaps in First and Third Party Evaluations

Foundation models are increasingly central to high-stakes AI systems, and governance frameworks now depend on evaluations to assess their risks and capabilities. Although general capability evaluations are widespread, social impact assessments covering bias, fairness, privacy, environmental costs, and labor practices remain uneven across the AI ecosystem. To characterize this landscape, we conduct the first comprehensive analysis of both first-party and third-party social impact evaluation reporting across a wide range of model developers. Our study examines 186 first-party release reports and 183 post-release evaluation sources, and complements this quantitative analysis with interviews of model developers. We find a clear division of evaluation labor: first-party reporting is sparse, often superficial, and has declined over time in key areas such as environmental impact and bias, while third-party evaluators including academic researchers, nonprofits, and independent organizations provide broader and more rigorous coverage of bias, harmful content, and performance disparities. However, this complementarity has limits. Only model developers can authoritatively report on data provenance, content moderation labor, financial costs, and training infrastructure, yet interviews reveal that these disclosures are often deprioritized unless tied to product adoption or regulatory compliance. Our findings indicate that current evaluation practices leave major gaps in assessing AI's societal impacts, highlighting the urgent need for policies that promote developer transparency, strengthen independent evaluation ecosystems, and create shared infrastructure to aggregate and compare third-party evaluations in a consistent and accessible way.

Racial/Ethnic Categories in AI and Algorithmic Fairness: Why They Matter and What They Represent

Racial diversity has become increasingly discussed within the AI and algorithmic fairness literature, yet little attention is focused on justifying the choices of racial categories and understanding how people are racialized into these chosen racial categories. Even less attention is given to how racial categories shift and how the racialization process changes depending on the context of a dataset or model. An unclear understanding of \textit{who} comprises the racial categories chosen and \textit{how} people are racialized into these categories can lead to varying interpretations of these categories. These varying interpretations can lead to harm when the understanding of racial categories and the racialization process is misaligned from the actual racialization process and racial categories used. Harm can also arise if the racialization process and racial categories used are irrelevant or do not exist in the context they are applied. In this paper, we make two contributions. First, we demonstrate how racial categories with unclear assumptions and little justification can lead to varying datasets that poorly represent groups obfuscated or unrepresented by the given racial categories and models that perform poorly on these groups. Second, we develop a framework, CIRCSheets, for documenting the choices and assumptions in choosing racial categories and the process of racialization into these categories to facilitate transparency in understanding the processes and assumptions made by dataset or model developers when selecting or using these racial categories.