Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AI Governance and Accountability: An Analysis of Anthropic's Claude

Aman Priyanshu, Yash Maurya, Zuofei Hong

TL;DR

The paper addresses governance and accountability challenges for high-impact LLMs, focusing on Anthropic's Claude. It applies the NIST AI Risk Management Framework and the EU AI Act to map threats and evaluate governance controls, including Anthropic's Constitutional AI. Key findings include transparency gaps in privacy policies, risks of hallucinations and biases, and data-usage concerns in partnerships. The authors propose mitigations—transparent data practices, rigorous benchmarking, and a remediation pipeline for data deletion and unlearning—that aim to enhance trust and guide responsible deployment.

Abstract

As AI systems become increasingly prevalent and impactful, the need for effective AI governance and accountability measures is paramount. This paper examines the AI governance landscape, focusing on Anthropic's Claude, a foundational AI model. We analyze Claude through the lens of the NIST AI Risk Management Framework and the EU AI Act, identifying potential threats and proposing mitigation strategies. The paper highlights the importance of transparency, rigorous benchmarking, and comprehensive data handling processes in ensuring the responsible development and deployment of AI systems. We conclude by discussing the social impact of AI governance and the ethical considerations surrounding AI accountability.

AI Governance and Accountability: An Analysis of Anthropic's Claude

TL;DR

The paper addresses governance and accountability challenges for high-impact LLMs, focusing on Anthropic's Claude. It applies the NIST AI Risk Management Framework and the EU AI Act to map threats and evaluate governance controls, including Anthropic's Constitutional AI. Key findings include transparency gaps in privacy policies, risks of hallucinations and biases, and data-usage concerns in partnerships. The authors propose mitigations—transparent data practices, rigorous benchmarking, and a remediation pipeline for data deletion and unlearning—that aim to enhance trust and guide responsible deployment.

Abstract

As AI systems become increasingly prevalent and impactful, the need for effective AI governance and accountability measures is paramount. This paper examines the AI governance landscape, focusing on Anthropic's Claude, a foundational AI model. We analyze Claude through the lens of the NIST AI Risk Management Framework and the EU AI Act, identifying potential threats and proposing mitigation strategies. The paper highlights the importance of transparency, rigorous benchmarking, and comprehensive data handling processes in ensuring the responsible development and deployment of AI systems. We conclude by discussing the social impact of AI governance and the ethical considerations surrounding AI accountability.
Paper Structure (31 sections, 9 figures)

This paper contains 31 sections, 9 figures.

Table of Contents

  1. Introduction
  2. Organization of this Report
  3. Literature Review
  4. Preliminaries
  5. Constitutional AI
  6. Threat Analysis
  7. Identified Issues
  8. Constitutional AI
  9. NIST Framework Analysis
  10. NIST "Govern" (Governance Analysis)
  11. NIST "MAP"
  12. NIST "Manage"
  13. NIST "MEASURE"
  14. EU AI Act Analysis
  15. Proposed Mitigations & Resolution Strategies
  16. ...and 16 more sections

Figures (9)

  • Figure 1: Anthropic's Claude is one of the most popular large language model chatbots available to the everyday consumer. This paper presents a study of its practices and conduct through the lens of AI governance.
  • Figure 2: Rapidly growing customer visits on their Claude's web interface
  • Figure 3: Some of Anthropic's Partnerships
  • Figure 4: Anthropic's Constitutional AI training process
  • Figure 5: "BBQparrish-etal-2022-bbq bias scores. Higher scores indicate more negative stereotype bias (lower is better). We used the same methods, code, and controls from our previously published work. The Public model shows lower bias scores across all nine social dimensions than the Standard model, especially for Disability Status and Physical Appearance. The Public constitution places a larger emphasis on accessibility, which may explain the greater reduction in bias for Disability Status in particular."CollectiveConstitution
  • ...and 4 more figures