Return to Articles

Getting started with AI trust and safety

Technical Report
Screen shot of the User Guide website.

Introducing The AI Alliance Trust and Safety User Guide, now available here: the-ai-alliance.github.io/trust-safety-user-guide/

This “living” document provides an introduction to current trends in research and development for ensuring AI models and applications meet requirements for trustworthy results, and in particular, results that satisfy various safety criteria. Aimed at developers and leaders who are relatively new to this topic, the guide defines some common terms, provides an overview of several leading trust and safety education and technology projects, and offers recommendations for how to build-in trust and safety into your AI-based applications.

The leading trust and safety projects discussed include the Risk Management Framework from the National Institute of Standards and Technology (NIST), Trust and Safety at Meta, The Mozilla Foundation’s guidance on Trustworthy AI, The MLCommons Taxonomy of Hazards, and others. 

We welcome your contributions! 

We intend to evolve this living document, in collaboration with the broader AI community, to reflect trends in trust and safety, and to provide more in-depth guidance and usable examples. The guide is published using GitHub Pages, allowing anyone to contribute improvements as pull requests in the guide source repo.

Contribute to the AI Trust and Safety guide

Related Articles

View All

Architecture of Data Prep Kit Framework 

Technical Report

The Data Prep Kit (DPK) framework enables scalable data transformation using Python, Ray, and Spark, while supporting various data sources such as local disk, S3, and Hugging Face datasets. It defines abstract base classes for transformations, allowing developers to implement custom data and folder transforms that operate seamlessly across different runtimes. DPK also introduces a data abstraction layer to streamline data access and facilitate checkpointing. To support large-scale processing, it provides three runtimes: Python for small datasets, Ray for distributed execution across clusters, and Spark for highly scalable processing using Resilient Distributed Datasets (RDDs). Additionally, DPK integrates with Kubeflow Pipelines (KFP) for automating transformations within Kubernetes environments. The framework includes transform utilities, testing support, and simplified APIs for invoking transforms efficiently. By abstracting complexity, DPK simplifies development, deployment, and execution of data processing pipelines in both local and distributed environments.

Advancing Domain-Specific Q&A: The AI Alliance's Guide to Best Practices

Technical Report

The AI Alliance application and tools working group has conducted a comprehensive study on best practices for advancing domain-specific Q&A using retrieval-augmented generation (RAG) techniques. The findings of this research, provide insights and recommendations for maximizing the capabilities of Q&A AI in specialized domains.

The State of Open Source AI Trust and Safety - End of 2024 Edition

News

We conducted a survey with 100 AI Alliance members to learn about the state of open source AI trust and safety for 2024. This blog post highlights key findings on AI applications, model popularity, safety concerns, regulatory focus, and gaps in current safety practices, while also providing an overview of notable open-source projects, tools, and research in the field of AI trust and safety.