Particle.news
Download on the App Store

Microsoft Open‑Sources RAMPART and Clarity to Embed AI Safety in Development

Microsoft says the tools convert red‑team results into repeatable tests to move AI safety into everyday engineering practice.

Overview

  • Microsoft released two open‑source projects, RAMPART and Clarity, publicly on May 20 to help developers build and test agentic AI earlier in the lifecycle.
  • RAMPART is a pytest‑native test framework that runs adversarial and benign scenarios, focuses on cross‑prompt injection attacks, and supports repeated statistical trials to handle non‑deterministic model behavior.
  • Clarity guides teams through problem definition, solution options, failure analysis, and decision tracking and writes human‑readable outputs to a .clarity‑protocol folder in the repo.
  • Microsoft says it already used RAMPART in incident response to reproduce vulnerabilities, generate variants, and shorten weeks of manual work to hours, and intends CI integration so tests can block regressions.
  • By open‑sourcing both tools Microsoft aims to scale red‑team lessons into living artifacts, encourage community contributions, and push organizations to treat agent risks like prompt injection and unintended tool use as engineering problems.