These tools and metrics are designed to help AI actors develop and use trustworthy AI systems and applications that respect human rights and are fair, transparent, explainable, robust, secure and safe.
The AI Safety Benchmark v0.5
The AI Safety Benchmark v0.5 is a benchmark for assessing the safety risks of AI systems that use chat-tuned language models. It was created by the MLCommons AI Safety Working Group and published in April 2024. The authors released it as a proof of concept, and state that it should not be used to assess the safety of AI systems. Its purpose is to set out a principled approach to building such benchmarks, ahead of a full version 1.0.
Version 0.5 covers a single use case: an adult chatting in English with a general-purpose assistant. It considers three types of user: typical, malicious and vulnerable. The benchmark introduces a new taxonomy of 13 hazard categories, with definitions and subcategories. Seven of these have tests in v0.5, including violent crimes, hate and suicide and self-harm.
In total, the tests contain 43,090 prompts, created using templates. AI systems are run on these prompts, and their responses are assessed for safety. A grading system then rates each system against the benchmark. The release also includes ModelBench, an openly available platform and downloadable tool for evaluating AI systems on the benchmark. An example evaluation report covers more than a dozen openly available chat-tuned language models, and a test specification documents the approach. The authors document the benchmark's limitations, flaws and challenges in detail.
Auto-discovered on 2026-02-11 by OECD Catalogue Automation
About the tool
You can click on the links to see the associated tools
Tool type(s):
Objective(s):
Country/Territory of origin:
Lifecycle stage(s):
Type of approach:
Maturity:
Usage rights:
License:
Target users:
Risk management stage(s):
Technology platforms:
Programming languages:
Use Cases
Would you like to submit a use case for this tool?
If you have used this tool, we would love to know more about your experience.
Add use case
Partnership on AI
