# Research that makes governance examinable. Read foundational work on documentation, evaluation, accountability and systemic risk. ## Use research with context Retain author, publication venue, version, source and retrieval provenance. Separate a paper’s findings from Alpha interpretation and from claims about a particular organization. ## Independent editorial review Research candidates require source and rights review, reviewer attribution, conflict disclosure and revision history before publication as Alpha conclusions. Sponsorship or Council membership does not control findings. ## Model Cards for Model Reporting Documents intended uses, evaluation conditions and limitations of trained models. Source: https://arxiv.org/abs/1810.03993v2 Margaret Mitchell et al.; version v2; publication unknown; editorial review pending. Citation and original summary only; no full-text ingestion ## Datasheets for Datasets Proposes structured documentation of dataset motivation, composition, collection and intended use. Source: https://arxiv.org/abs/1803.09010 Timnit Gebru et al.; version unknown; publication unknown; editorial review pending. Citation and original summary only; no full-text ingestion ## Closing the AI Accountability Gap Presents an end-to-end framework for internal algorithmic auditing. Source: https://arxiv.org/abs/2001.00973 Inioluwa Deborah Raji et al.; version v1; publication 2020-01-03; editorial review pending. Citation and original summary only; no full-text ingestion ## On the Dangers of Stochastic Parrots Examines environmental, financial and societal concerns associated with large language models. Source: https://doi.org/10.1145/3442188.3445922 Emily M. Bender et al.; version FAccT 2021; publication 2021-03-01; editorial review pending. Citation and original summary only; no full-text ingestion ## Holistic Evaluation of Language Models Examines language models across multiple scenarios and metrics; the linked revision is dated 2023-10-01. Source: https://arxiv.org/abs/2211.09110v2 Percy Liang et al.; version v2; publication 2022-11-16; editorial review pending. Citation and original summary only; no full-text ingestion ## The AI Risk Repository A research entry point to risk classifications and the associated meta-review. Source: https://airisk.mit.edu/ Peter Slattery et al.; version unknown; publication unknown; editorial review pending. Citation and original summary only; no full-text ingestion ## The Foundation Model Transparency Index The foundational 2023 paper. Later editions exist; do not treat historical findings as current developer assessments. Source: https://arxiv.org/abs/2310.12941v1 Rishi Bommasani et al.; version v1; publication 2023-10-19; editorial review pending. Citation and original summary only; no full-text ingestion ## Responsible Practices for Synthetic Media A framework for responsible development, creation and sharing of synthetic media, organized around consent, disclosure and transparency. Source: https://syntheticmedia.partnershiponai.org/ Partnership on AI; version unknown; publication 2023-02-22; editorial review pending. Citation and original summary only; no full-text ingestion ## Measuring AI Ability to Complete Long Software Tasks Research on measuring agent capability through task duration and reliability. Its benchmark results do not establish safety or performance in a particular enterprise deployment. Source: https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/ METR; version unknown; publication 2025-03-19; editorial review pending. Citation and original summary only; no full-text ingestion ## Project Moonshot Primary project documentation for an open-source testing toolkit covering LLM application benchmarks and red teaming. Tool availability is not an assurance conclusion. Source: https://aiverifyfoundation.sg/tools/moonshot/ AI Verify Foundation; version unknown; publication unknown; editorial review pending. Citation and original summary only; no full-text ingestion ## Algorithmic accountability for the public sector Research on early algorithmic accountability policies in the public sector. Scope and historical context matter when applying its findings. Source: https://www.adalovelaceinstitute.org/report/algorithmic-accountability-public-sector/ Ada Lovelace Institute, AI Now Institute and Open Government Partnership; version unknown; publication 2021-08-24; editorial review pending. Citation and original summary only; no full-text ingestion ## Artificial Power: 2025 Landscape Report An institutional research report examining power and public agency in the AI landscape. The report’s analysis remains attributable to its authors. Source: https://ainowinstitute.org/publications/research/ai-now-2025-landscape-report AI Now Institute; version unknown; publication 2025-06-03; editorial review pending. Citation and original summary only; no full-text ingestion ## Responsible AI Institute research and governance resources Discovery entry for the Institute’s public research and governance resources. Member-only material is excluded; this entry is not a specific publication or an Alpha certification claim. Source: https://www.responsible.ai/ Responsible AI Institute; version unknown; publication unknown; editorial review pending. Citation and original summary only; no full-text ingestion ## Explore - [Responsible AI Observatory](/responsible-ai) - [Standards inform governance. Scope determines applicability.](/responsible-ai/standards) - [Research that makes governance examinable.](/responsible-ai/research) - [Make risk accountable.](/responsible-ai/risks) - [Build judgment before the consequential decision.](/responsible-ai/board-education) - [Make the method visible.](/responsible-ai/methodology) - [Comparable evidence comes before comparison.](/responsible-ai/benchmarks)