Microsoft AI Updates: May 24, 2026
1. GitHub Copilot CLI v1.0.52 tightens autopilot, security review, and session resume
Microsoft. GitHub shipped Copilot CLI v1.0.52 with 24 fixes and improvements aimed at sharper terminal ergonomics, including a new /security-review slash command for vulnerability scanning and a /usage view that surfaces quota progress bars and AI Credits budget management. Sessions now resume in their saved working directory with refreshed git context, the autopilot mode stops triggering unexpected permission prompts, and general-purpose subagents automatically use GPT-5.4 or GPT-5.5 when available. The release also enforces 200K versus 1M context-window tier selection end-to-end and reports reasoning tokens separately in usage summaries. Source
2. Foundry IQ pitched as a shared knowledge brain across multiple agents
Microsoft. Microsoft’s Azure AI Foundry team detailed how Foundry IQ acts as a centralized, multi-source knowledge layer that several agents can share, replacing per-agent retrieval logic with reusable, permission-aware knowledge bases. The service can index SharePoint, OneLake, Fabric IQ, Azure Blob Storage, Azure AI Search, the web, and MCP sources (in private preview), and runs an agentic retrieval engine that plans, decomposes, and iterates on queries before returning cited context to the model. The post positions Foundry IQ as the cross-agent counterpart to Work IQ inside Microsoft’s IQ Series. Source
3. Microsoft Foundry observability stack centralizes evals, traces, and monitoring in Azure Monitor
Microsoft. A new Foundry blog walked through the now generally available evaluation, monitoring, and tracing capabilities, all published directly into Azure Monitor so agent telemetry shares an operational plane with the rest of the infrastructure stack. Built-in evaluators cover quality (coherence, fluency), RAG (groundedness, relevance), safety (hate, violence, protected materials), and agent metrics like tool-call accuracy and task completion, while the Agent Monitoring Dashboard tracks token usage, latency, success rates, and evaluation outcomes on production traffic. Microsoft frames the suite as the way to triage groundedness regressions in minutes by separating model, retrieval, and infrastructure causes. Source