Safety
Focuses on proactive prevention and mitigation of harms, with a focus on content and conduct safety for generated and agentic outputs for “everyday” AI products and non-frontier models, as well as frontier safety concerns applicable to API availability and the integration of frontier models into consumer-facing products. This work complements that of the security working group, which focuses on malicious attacks by bad actors and privacy concerns, and informs all AIRR work streams.
Purpose
The world is rapidly moving from AI safety frameworks and best practices, toward international standards. The AIRR safety workstream builds upon x-industry responsible AI governance strategies and trust and safety practices to complement all other AIRR working groups by focusing acutely on the new risks of AI-generated content and relevant frontier safety and emerging agentic safety concerns. These risks are distinct from – and their mitigations supplement those for – privacy harms and security (i.e., malicious user, cyber) threats. The Safety workstream focuses on evolving product policies to govern and mitigate these risks. The workstream is informed by ML Commons multiregional safety research by the Multimodal work stream. The group focuses on how to provide consistent technical benchmarking to inform and meet evolving international standards related to AI-generated and agentic content and conduct safety and policies.
Deliverables
2026 Goals for the working group include:
Achieve expert consensus on AI safety definitions and technical edits pre ISO/ IEC SC 42 Plenary in October for specific drafts, e.g., 42119 series (testing); WG3 PWI on AI Safety and Trustworthiness Fact Labels, to support AI Reliability Label as a means for ISO IEC (and other) compliance
Publish a blog post or other thought leadership for a non-technical audience (e.g., policy, business decision-maker, mainstream) explaining what AI safety benchmarks are, how they are developed, why we need more contributors, and why they are important as the world moves from safety frameworks toward benchmarks – could be World Economic Forum Agenda Blog, Tech Policy Press, or other broad audience
Offer clear utility for integration of reliability labels for multi stakeholder initiatives for AI Safety as standards develop, e.g., Appia, Partnership on AI Hub, UN Global Dialogue for AI Governance, etc.
Paper offering a baseline framework for building AI safety benchmark datasets – for submission to international conference and/or journal (e.g., FAccT)
NeurIPS 2026 workshop
Meeting Schedule
Wednesdays Monthly on the fourth Wednesday, 1pm – 2pm ET
AI Risk & Reliability Working Group Projects
How to Join and Access Resources
To sign up for the group mailing list and receive the meeting invite:
- Fill out our subscription form and indicate that you’d like to join the AI Risk & Reliability Working Group.
- Associate a Google account with your organizational email address.
- Once your request to join the AI Risk & Reliability working group is approved, you’ll be able to access the AI Risk & Reliability folder in the Public Google Drive.
To access the GitHub repositories (public):
- If you want to contribute code, please submit your GitHub username to our subscription form.
- Visit the GitHub repositories:
Working Group Chairs
William Bartholomew
William works at the intersection of responsible AI, cybersecurity, software supply chains, open source, and standards development.
As a Director of Public Policy in Microsoft’s Office of Responsible AI, William helps shape policy approaches for advanced AI systems, including foundation models, copilots, and AI agents. His work focuses on making responsible AI governance technically grounded, operationally feasible, and scalable across products, organizations, and jurisdictions.
William’s policy work is informed by more than a decade designing, implementing, and operating software supply chains used by tens of thousands of developers. He brings a practical, friction-reducing mindset to standards development, open source, and public- and private-sector working groups globally.
Reena Jana
Reena is the Head of AI Safety Ecosystem Strategy at Google. Within the Trust & Safety organization, she leads a broad, global portfolio of external engagements with trusted organizations to build a collaborative, ecosystem approach to the research and development of best practices and standards for AI safety. Reena serves as a senior expert and advisor on AI product safety and governance.
Reena is also engaged with numerous cross-industry efforts on best practices and the development of shared practices, with multi-stakeholder organizations including the World Economic Forum, the Ad Council, Partnership on AI, US Telecommunications Training Institute.