Observed Signal · Jun 2, 2026 · Technical Release · Source: techcrunch · Impact: 4/5 · Sentiment: Positive
Microsoft releases ASSERT for AI behavior testing
Microsoft announced ASSERT (Adaptive Spec-driven Scoring for Evaluation and Regression Testing), an open-source framework that converts high-level, natural-language descriptions of desired behaviors and policies into structured, scored tests for application-specific AI evaluation. ASSERT generates acceptable/unacceptable behavior specs, creates problem scenarios and test cases, runs them against target systems, and records execution traces including intermediate actions and tool calls for investigation. Developers can supply system context, tools, and constraints to customize evaluations. Microsoft positions ASSERT as filling a gap left by broader benchmarks and recommends using it during development, after deployment, and for continuous monitoring. The release is contextualized alongside broader evaluation efforts such as Stanford’s HELM, MLCommons’ AILuminate, and evaluation groups like METR.
A major platform (Microsoft) released an open-source technical framework for application-specific AI behavior evaluation and continuous monitoring; this affects how developers validate, govern, and deploy AI systems across products, improving trust, compliance, and operational testing practices.
Track Microsoft Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Microsoft released ASSERT (Adaptive Spec-driven Scoring for Evaluation and Regression Testing) as an open-source framework.
- ASSERT transforms plain-language descriptions of expected AI behavior into structured sets of acceptable and unacceptable behaviors and generates test cases.
- The framework can run generated tests against a target system and score results, while recording execution paths, intermediate actions and tool calls for debugging.
- ASSERT is published on GitHub and is intended for use during development, post-deployment checks, and continuous monitoring of AI systems.
Connected Companies & Entities
2 Entities mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Microsoft launches Agent Control Specification for AI
Microsoft has published an open source standard called the Agent Control Specification (ACS) to give developers, security, and compliance teams a consistent, granular way to govern AI agents. ACS lets teams write policy files that declare what an agent may or must not do, when human approval is required, and what evidence to log. Policies are evaluated at multiple interception points across an agent workflow (before input, before tool calls, after tool returns, and before final response). The spec supports inserting classifiers, LLM-based policy judges, and tool-call logic, and policies can be bundled with agents for portability across frameworks. Microsoft is shipping ACS as an SDK with plug-ins for LangChain, OpenAI Agents SDK, Anthropic Agents SDK, AutoGen, CrewAI, Semantic Kernel, Microsoft.Extensions.AI, MCP tools, and others.
Microsoft publishes AI code of conduct restricting cyberattacks and deception.
Microsoft has released a new AI code of conduct, focusing on safety and alignment principles for its AI models. The document outlines values and red lines that guide model training within Microsoft AI, including absolute constraints against cyberattacks, nuclear weapons, and deepfake production. It also emphasizes that models must support humans rather than replace them and must not use deceptive or self-reinforcing mechanisms to evade human oversight. The release comes amid heightened AI safety concerns and industry-wide discussions on pacing the frontier. Microsoft CEO Satya Nadella expressed support for ideas like embedded evaluators to ensure alignment.
Most Developers Test Code — Why Not Test AI?
The article argues that AI features need the same rigorous testing workflows as traditional software. Developers commonly rely on informal manual checks for AI outputs (e.g., "I tried it three times and it seems pretty good"), but AI components (retrieval, prompts, LLMs, validation) can fail in many ways and are probabilistic. The author recommends building small evaluation datasets (20–50 representative test cases), testing prompts, context, and end-to-end workflows, and integrating evaluation into CI/GitHub workflows to measure whether changes (prompts, models, retrieval) actually improve performance. The piece presents a three-layer evaluation rule—produce an answer, produce correct answers consistently, and be able to measure improvement—and calls for treating evaluation as a first-class engineering concern.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.