Responsible AI Collaborative

The Responsible AI Collaborative (RAIC) is a U.S. nonprofit (EIN 88-1046583) that operates the AI Incident Database (AIID) — the primary public repository tracking real-world failures and harmful outcomes from deployed AI systems. Founded in 2022, the RAIC institutionalized a project that machine learning safety researcher Sean McGregor had launched in November 2020, modeled explicitly on the incident-reporting databases that mature safety-critical industries — aviation, nuclear, cybersecurity — use to prevent repeated failures. The AIID occupies a distinctive position in the movement: it is shared infrastructure that civil society organizations, journalists, litigators, policymakers, and researchers all draw on to document and investigate AI harms, making the RAIC an organizational anchor for the broader algorithmic accountability ecosystem.

Origins and founding

Sean McGregor launched the AI Incident Database publicly in November 2020 and published its foundational rationale as "Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database" (IAAI-21, 2020). The paper's core argument was that AI deployment lacks the collective institutional memory that makes mature industrial sectors safer: aviation has the NTSB, cybersecurity has CVE and NVD, nuclear power has the IAEA's incident reporting regime — but developers of AI systems had no shared mechanism for learning from documented failures, causing the same classes of harm to recur independently across organizations. The database launched with more than 1,000 archived incident reports, drawn from news coverage and technical documentation of AI harms across domains including autonomous vehicle incidents, algorithmic trading failures, and facial recognition wrongful arrests. The Partnership on AI served as database founding sponsor. The Responsible AI Collaborative was formally incorporated in 2022, with the Wu Foundation providing organizational founding support; hiring for founding staff was announced in March 2022 and the organization held its first all-hands meeting in July 2022.

The AI Incident Database

The AIID provides full-text and faceted search across a continually growing corpus of AI incident reports. Each incident links to source documentation and is classified using structured taxonomies that enable researchers and policymakers to identify patterns across failure types, deployment contexts, and affected populations. The database is community-supported: more than 127 contributors have submitted incidents, ranging from affiliated researchers to independent monitors. Daniel Atherton, the RAIC's lead editor, has made the most substantial contribution — 943 new incidents and more than 1,000 additional reports. The submission pipeline is public and open, making the AIID one of the few AI-accountability tools whose data actively depends on civil society and researcher participation rather than internal production alone.

Three taxonomy systems currently structure the database:

• CSETv1 (Center for Security and Emerging Technology) — classifies incidents by harm distribution basis and sector of deployment for the policy community
• GMF (Goals, Methods, and Failures) — a technical framework examining AI objectives, technologies, and procedural breakdowns that contributed to incidents
• MIT AI Risk Repository — a 23-domain risk classification system covering discrimination, misinformation, and other harm categories

These overlapping frameworks reflect the AIID's positioning as shared infrastructure: different communities of users — technical researchers, policy analysts, civil rights advocates — access the same incident corpus through different analytical lenses.

Governance and funding

The RAIC is governed by a voting board including Patrick Hall (George Washington School of Business), Heather Frase (Veraitech; Virginia Tech's National Security Institute), and Kristian J. Hammond (Northwestern University; founder of Narrative Science). Sean McGregor and Helen Toner (Georgetown's Center for Security and Emerging Technology) serve as emeritus board members. McGregor, who built the AIID as a machine learning safety researcher, also leads the ML Commons Agentic Workstream and founded the Digital Safety Research Institute at UL Research Institutes. Funders include the MacArthur Foundation (grants), the Morningside Technology Foundation, and in-kind support from Netlify, Cloudinary, Algolia, and Trustible.

Role in the accountability movement

The AIID functions as movement infrastructure rather than an advocacy organization: it publishes the evidence base that other actors draw on. Journalists investigating specific AI harms can search the database for prior documented incidents in the same domain. Civil rights litigators can use incident taxonomies to establish patterns of algorithmic discrimination. Regulators can trace harm clusters across deployment sectors. Policymakers designing AI incident-reporting requirements — a growing area of AI governance legislation globally — have a template in the AIID's architecture. The RAIC also publishes a quarterly incident roundup and operates a news digest that surfaces AI-related reporting for database cross-referencing. Its 2026 Africa-focused work — examining AI safety incidents across the continent and building regional incident-monitoring capacity — extends the registry beyond its early U.S. and European coverage patterns.