Recommendation to Establish the California AI Accountability and Redress Act
Daniel Vennemeyer, Phan Anh Duong · Team Taylor Avenue
Submitted to Berkeley AI Policy Hackathon. Projects from partner hackathons are early-stage work by participants, not Apart Research publications.
We recommend that California enact the California AI Accountability and Redress Act (CAARA) to address emerging harms from the widespread deployment of generative AI, particularly large language models (LLMs), in public and commercial systems.
This policy would create the California AI Incident and Risk Registry (CAIRR) under the California Department of Technology to transparently track post-deployment failures. CAIRR would classify model incidents by severity, triggering remedies when a system meets the criteria for an "Unacceptable Failure" (e.g., one Critical incident or three unresolved Moderate incidents).
Critically, CAARA also protects developers and deployers by: Setting clear thresholds for liability; Requiring users to prove demonstrable harm and document reasonable safeguards (e.g., disclosures, context-sensitive filtering); and Establishing a safe harbor for those who self-report and remediate in good faith.
CAARA reduces unchecked harm, limits arbitrary liability, and affirms California’s leadership in AI accountability.
Reviews
No public critique yet.
Cite this project
@misc{vennemeyer2025recommendation,
title = {{Recommendation to Establish the California AI Accountability and Redress Act}},
author = {Daniel Vennemeyer and Phan Anh Duong},
year = {2025},
month = apr,
note = {Submitted to Berkeley AI Policy Hackathon, a partner hackathon},
howpublished = {\url{https://apartresearch.com/sprints/projects/recommendation-to-establish-the-california-ai-accountability-and-redress-act-buqi}},
url = {https://apartresearch.com/sprints/projects/recommendation-to-establish-the-california-ai-accountability-and-redress-act-buqi}
}More from Berkeley AI Policy Hackathon
- View project: Recursive Fitness Alignment Protocol (RFAP)
Recursive Fitness Alignment Protocol (RFAP)
CC4CI — Caribbean Center for Collective Intelligence
The Recursive Fitness Alignment Protocol (RFAP) is a minimal testbed for aligning reasoning systems—AI, human, or institutional—through recursive self-correction. Rather than optimizing for fixed goals or architectures, …
- View project: Round 1 Submission
Round 1 Submission
Rand#1
Our round one submission for the UC Berkeley AI Policy Hackathon
- View project: CalSandbox and the CLEAR AI Act: A State-Led Vision for Responsible AI Governance
CalSandbox and the CLEAR AI Act: A State-Led Vision for Responsible AI Governance
Kalani & Gomes
The CLEAR AI Act proposes a balanced, forward-looking framework for AI governance in California that protects public safety without stifling innovation. Centered around CalSandbox, a supervised testbed for high-risk AI …