NewsAgency

Microsoft Releases Open-Source ASSERT Framework for AI Agent Behavior Testing

Properties1
tagsmicrosoft, assert, ai-testing, build-2026, agent-safety

3 dakika okuma süresi

Summary

Microsoft unveiled ASSERT (Adaptive Spec-driven Scoring for Evaluation and Regression Testing) at Build 2026 on June 2, 2026 — an open-source MIT-licensed framework that converts natural-language policy descriptions into structured, scored test cases for AI agents. ASSERT generates acceptable/unacceptable behavior categories, problem scenarios, and multi-turn tests; records agent execution paths including tool calls; and supports continuous monitoring post-deployment. It is model-agnostic, integrates with LangChain, CrewAI, LiteLLM, and OpenAI via OpenTelemetry/OpenInference tracing, and complements the Agent Control Specification (ACS) in Microsoft’s agent trust stack.

PreScreening Notes

Score: 7/10 (High) — Practical open-source developer tool for AI agent testing, announced at Build 2026. TechCrunch is credible, recent (June 2), strong fit for software/AI audience building agentic systems. No duplicate in pipeline.

Source Analysis

Research Notes

Additional Sources

Key Facts Verified

  • Confirmed: MIT license; natural-language-to-test-case conversion
  • Confirmed: LangChain, CrewAI, LiteLLM, OpenAI integration via OpenTelemetry
  • Confirmed: Complements Agent Control Specification (ACS)

Broader Context

Practical open-source tooling for agent regression testing at Build 2026.

assert-framework, microsoft, microsoft-agent-framework, ai-safety, build-2026

Draft Article

Microsoft ASSERT: Doğal Dille AI Agent Davranış Testi — MIT Lisanslı Açık Kaynak

microsoft, build-2026’da ASSERT (Adaptive Spec-driven Scoring for Evaluation and Regression Testing) framework’ünü duyurdu. MIT lisanslı açık kaynak araç, doğal dil policy tanımlarını yapılandırılmış test case’lere dönüştürüyor. Model-agnostic; LangChain, CrewAI, LiteLLM ve OpenAI entegrasyonu OpenTelemetry tracing ile.

Ana Gelişme

İşlev: Natural-language → acceptable/unacceptable behavior kategorileri, problem senaryoları, multi-turn testler
Tracing: Agent execution path ve tool call kaydı
Monitoring: Post-deployment continuous monitoring
Tamamlayıcı: Agent Control Specification (ACS) — Microsoft agent trust stack’inin parçası

Neden Önemli?

ai-safety ve agent builder’lar için pratik regression testing aracı. Türk startup ekosisteminde agentic sistem geliştiren ekipler hemen adopt edebilir — MAI-Thinking-1 duyurusundan ayrı Build 2026 cluster öğesi.

Bağlam

microsoft-agent-framework ve assert-framework birlikte Microsoft’un agent governance stratejisini oluşturuyor.

Kaynaklar

Evaluation Report

News Value Assessment

  • Timeliness: Build 2026, June 2, 2026.
  • Impact: Open-source MIT framework for AI agent behavior testing — practical developer tooling.
  • Prominence: Microsoft, TechCrunch coverage.
  • Proximity: High — agent builders in Turkish startup ecosystem can adopt immediately.
  • Novelty: Natural-language-to-test-case conversion for agent regression testing.

Audience Fit

  • Primary software developer audience building agentic systems.

Risk & Ethics Assessment

  • Verification: TechCrunch + Microsoft Build announcement. Low risk.
  • Can be published alongside MAI-Thinking-1 as Build 2026 cluster but distinct angle.

Publication Strategy

Suggested Angle

ASSERT: Doğal dilden AI agent test senaryosu — LangChain/CrewAI entegrasyonu ve Türk geliştiriciler için hızlı başlangıç rehberi.

Editorial Notes

Status: Approved — practical developer tool
Format: brief
Angle: Confirmed — doğal dilden agent test senaryosu

Headline Suggestions (Turkish)

  • Microsoft ASSERT: Doğal dille AI agent davranış testi — MIT lisanslı açık kaynak
  • ASSERT framework: LangChain ve CrewAI entegrasyonu Build 2026’da
  • AI agent regression testing: Microsoft’un Agent Control Specification tamamlayıcısı

Key Points (Must Include)

  • MIT lisans; natural-language → structured test cases
  • LangChain, CrewAI, LiteLLM, OpenAI; OpenTelemetry tracing
  • ACS (Agent Control Specification) tamamlayıcısı
  • Post-deployment continuous monitoring

Reporting Instructions

  • Brief format; MAI-Thinking-1 ile ayrı makale, Build 2026 cluster notu
  • Türk geliştiriciler için hızlı başlangıç bağlantıları

Grafik Görünümü