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

Anka Reuel, Avijit Ghosh, Jenny Chim, Andrew Tran, Yanan Long, Jennifer Mickel, Usman Gohar, Srishti Yadav, Pawan Sasanka Ammanamanchi, Mowafak Allaham, Hossein A. Rahmani, Mubashara Akhtar, Felix Friedrich, Robert Scholz, Michael Alexander Riegler, Jan Batzner, Eliya Habba, Arushi Saxena, Anastassia Kornilova, Kevin Wei, Prajna Soni, Yohan Mathew, Kevin Klyman, Jeba Sania, Subramanyam Sahoo, Olivia Beyer Bruvik, Pouya Sadeghi, Sujata Goswami, Angelina Wang, Yacine Jernite, Zeerak Talat, Stella Biderman, Mykel Kochenderfer, Sanmi Koyejo, Irene Solaiman
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:104906-104952, 2026.

Abstract

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 remain uneven. To characterize this landscape, we conduct the first comprehensive analysis of social impact evaluation reporting, examining 186 first-party release reports and 248 third-party evaluation sources, supplemented by developer interviews. We find a stark division of labor: first-party reporting is sparse, often superficial, and declining in areas like environmental impact and bias, while third-party evaluators provide broader, more rigorous coverage of bias, harmful content, and performance disparities. However, only developers can authoritatively report on data provenance, content moderation labor, costs, and infrastructure, yet interviews reveal these disclosures are deprioritized unless tied to product adoption or compliance. Current practices leave major gaps in assessing societal impacts, underscoring the need for policies that mandate developer transparency, strengthen independent evaluation ecosystems, and create shared infrastructure for aggregating third-party evaluations.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-reuel26a, title = {Who Evaluates {AI}’s Social Impacts? {M}apping Coverage and Gaps in First and Third Party Evaluations}, author = {Reuel, Anka and Ghosh, Avijit and Chim, Jenny and Tran, Andrew and Long, Yanan and Mickel, Jennifer and Gohar, Usman and Yadav, Srishti and Ammanamanchi, Pawan Sasanka and Allaham, Mowafak and Rahmani, Hossein A. and Akhtar, Mubashara and Friedrich, Felix and Scholz, Robert and Riegler, Michael Alexander and Batzner, Jan and Habba, Eliya and Saxena, Arushi and Kornilova, Anastassia and Wei, Kevin and Soni, Prajna and Mathew, Yohan and Klyman, Kevin and Sania, Jeba and Sahoo, Subramanyam and Bruvik, Olivia Beyer and Sadeghi, Pouya and Goswami, Sujata and Wang, Angelina and Jernite, Yacine and Talat, Zeerak and Biderman, Stella and Kochenderfer, Mykel and Koyejo, Sanmi and Solaiman, Irene}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {104906--104952}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/reuel26a/reuel26a.pdf}, url = {https://proceedings.mlr.press/v306/reuel26a.html}, abstract = {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 remain uneven. To characterize this landscape, we conduct the first comprehensive analysis of social impact evaluation reporting, examining 186 first-party release reports and 248 third-party evaluation sources, supplemented by developer interviews. We find a stark division of labor: first-party reporting is sparse, often superficial, and declining in areas like environmental impact and bias, while third-party evaluators provide broader, more rigorous coverage of bias, harmful content, and performance disparities. However, only developers can authoritatively report on data provenance, content moderation labor, costs, and infrastructure, yet interviews reveal these disclosures are deprioritized unless tied to product adoption or compliance. Current practices leave major gaps in assessing societal impacts, underscoring the need for policies that mandate developer transparency, strengthen independent evaluation ecosystems, and create shared infrastructure for aggregating third-party evaluations.} }
Endnote
%0 Conference Paper %T Who Evaluates AI’s Social Impacts? Mapping Coverage and Gaps in First and Third Party Evaluations %A Anka Reuel %A Avijit Ghosh %A Jenny Chim %A Andrew Tran %A Yanan Long %A Jennifer Mickel %A Usman Gohar %A Srishti Yadav %A Pawan Sasanka Ammanamanchi %A Mowafak Allaham %A Hossein A. Rahmani %A Mubashara Akhtar %A Felix Friedrich %A Robert Scholz %A Michael Alexander Riegler %A Jan Batzner %A Eliya Habba %A Arushi Saxena %A Anastassia Kornilova %A Kevin Wei %A Prajna Soni %A Yohan Mathew %A Kevin Klyman %A Jeba Sania %A Subramanyam Sahoo %A Olivia Beyer Bruvik %A Pouya Sadeghi %A Sujata Goswami %A Angelina Wang %A Yacine Jernite %A Zeerak Talat %A Stella Biderman %A Mykel Kochenderfer %A Sanmi Koyejo %A Irene Solaiman %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-reuel26a %I PMLR %P 104906--104952 %U https://proceedings.mlr.press/v306/reuel26a.html %V 306 %X 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 remain uneven. To characterize this landscape, we conduct the first comprehensive analysis of social impact evaluation reporting, examining 186 first-party release reports and 248 third-party evaluation sources, supplemented by developer interviews. We find a stark division of labor: first-party reporting is sparse, often superficial, and declining in areas like environmental impact and bias, while third-party evaluators provide broader, more rigorous coverage of bias, harmful content, and performance disparities. However, only developers can authoritatively report on data provenance, content moderation labor, costs, and infrastructure, yet interviews reveal these disclosures are deprioritized unless tied to product adoption or compliance. Current practices leave major gaps in assessing societal impacts, underscoring the need for policies that mandate developer transparency, strengthen independent evaluation ecosystems, and create shared infrastructure for aggregating third-party evaluations.
APA
Reuel, A., Ghosh, A., Chim, J., Tran, A., Long, Y., Mickel, J., Gohar, U., Yadav, S., Ammanamanchi, P.S., Allaham, M., Rahmani, H.A., Akhtar, M., Friedrich, F., Scholz, R., Riegler, M.A., Batzner, J., Habba, E., Saxena, A., Kornilova, A., Wei, K., Soni, P., Mathew, Y., Klyman, K., Sania, J., Sahoo, S., Bruvik, O.B., Sadeghi, P., Goswami, S., Wang, A., Jernite, Y., Talat, Z., Biderman, S., Kochenderfer, M., Koyejo, S. & Solaiman, I.. (2026). Who Evaluates AI’s Social Impacts? Mapping Coverage and Gaps in First and Third Party Evaluations. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:104906-104952 Available from https://proceedings.mlr.press/v306/reuel26a.html.

Related Material