AILuminate: Introducing v1.0 of the AI Risk and Reliability Benchmark from MLCommons
Abstract
The rapid advancement and deployment of AI systems have created an urgent need for standard safety-evaluation frameworks. This paper introduces AILuminate v1.0, the first comprehensive industry-standard benchmark for assessing AI-product risk and reliability. Its development employed an open process that included participants from multiple fields. The benchmark evaluates an AI system's resistance to prompts designed to elicit dangerous, illegal, or undesirable behavior in 12 hazard categories, including violent crimes, nonviolent crimes, sex-related crimes, child sexual exploitation, indiscriminate weapons, suicide and self-harm, intellectual property, privacy, defamation, hate, sexual content, and specialized advice (election, financial, health, legal). Our method incorporates a complete assessment standard, extensive prompt datasets, a novel evaluation framework, a grading and reporting system, and the technical as well as organizational infrastructure for long-term support and evolution. In particular, the benchmark employs an understandable five-tier grading scale (Poor to Excellent) and incorporates an innovative entropy-based system-response evaluation. In addition to unveiling the benchmark, this report also identifies limitations of our method and of building safety benchmarks generally, including evaluator uncertainty and the constraints of single-turn interactions. This work represents a crucial step toward establishing global standards for AI risk and reliability evaluation while acknowledging the need for continued development in areas such as multiturn interactions, multimodal understanding, coverage of additional languages, and emerging hazard categories. Our findings provide valuable insights for model developers, system integrators, and policymakers working to promote safer AI deployment.
Keywords
Cite
@article{arxiv.2503.05731,
title = {AILuminate: Introducing v1.0 of the AI Risk and Reliability Benchmark from MLCommons},
author = {Shaona Ghosh and Heather Frase and Adina Williams and Sarah Luger and Paul Röttger and Fazl Barez and Sean McGregor and Kenneth Fricklas and Mala Kumar and Quentin Feuillade--Montixi and Kurt Bollacker and Felix Friedrich and Ryan Tsang and Bertie Vidgen and Alicia Parrish and Chris Knotz and Eleonora Presani and Jonathan Bennion and Marisa Ferrara Boston and Mike Kuniavsky and Wiebke Hutiri and James Ezick and Malek Ben Salem and Rajat Sahay and Sujata Goswami and Usman Gohar and Ben Huang and Supheakmungkol Sarin and Elie Alhajjar and Canyu Chen and Roman Eng and Kashyap Ramanandula Manjusha and Virendra Mehta and Eileen Long and Murali Emani and Natan Vidra and Benjamin Rukundo and Abolfazl Shahbazi and Kongtao Chen and Rajat Ghosh and Vithursan Thangarasa and Pierre Peigné and Abhinav Singh and Max Bartolo and Satyapriya Krishna and Mubashara Akhtar and Rafael Gold and Cody Coleman and Luis Oala and Vassil Tashev and Joseph Marvin Imperial and Amy Russ and Sasidhar Kunapuli and Nicolas Miailhe and Julien Delaunay and Bhaktipriya Radharapu and Rajat Shinde and Tuesday and Debojyoti Dutta and Declan Grabb and Ananya Gangavarapu and Saurav Sahay and Agasthya Gangavarapu and Patrick Schramowski and Stephen Singam and Tom David and Xudong Han and Priyanka Mary Mammen and Tarunima Prabhakar and Venelin Kovatchev and Rebecca Weiss and Ahmed Ahmed and Kelvin N. Manyeki and Sandeep Madireddy and Foutse Khomh and Fedor Zhdanov and Joachim Baumann and Nina Vasan and Xianjun Yang and Carlos Mougn and Jibin Rajan Varghese and Hussain Chinoy and Seshakrishna Jitendar and Manil Maskey and Claire V. Hardgrove and Tianhao Li and Aakash Gupta and Emil Joswin and Yifan Mai and Shachi H Kumar and Cigdem Patlak and Kevin Lu and Vincent Alessi and Sree Bhargavi Balija and Chenhe Gu and Robert Sullivan and James Gealy and Matt Lavrisa and James Goel and Peter Mattson and Percy Liang and Joaquin Vanschoren},
journal= {arXiv preprint arXiv:2503.05731},
year = {2025}
}
Comments
51 pages, 8 figures and an appendix