Recent News in Analytics and AI: May 2026 Edition

If April was about confidence, May 2026 was about control. The most important updates did not just introduce new capabilities. They showed how analytics and AI platforms are being reshaped around trust, governance, reusable workflows and real enterprise execution. From Power BI and Fabric to Copilot, Foundry, Databricks and open-source projects, the common thread was a push to make advanced tools more dependable, connected and actionable. This edition highlights the updates that matter most for leaders trying to turn fast-moving analytics and AI innovation into something scalable, governed and genuinely useful.

Read on and get up to speed.

Power BI

• Power BI’s May 2026 update delivers a broad set of improvements across Copilot, reporting, modelling, and data connectivity, helping teams move faster from data ingestion to insight. Many feature reach the general availability milestone including visual calculations, custom totals now generally available, improved matrix behaviour, Input slicer numeric column support, and translytical task flow optional parameters. Why this matters: Copilot enhancements and version history strengthen productivity, while new visuals and formatting capabilities improve storytelling, making the platform more accessible and enterprise-ready. Learn more.

• The redesigned Power Query Get Data experience modernises how users discover and connect to data, introducing a streamlined interface that reduces friction throughout the workflow. Improved navigation, better connector discovery, and unified connection settings help users move from selecting data sources to shaping data more quickly. Why this matters: With added accessibility features such as keyboard navigation and dark mode, the update supports a broader range of users while aligning Power Query experiences across Fabric tools. Learn more.

• Outbound Access Protection introduces a new layer of governance by controlling where semantic models can send data. Enabled at workspace level, it blocks outbound connections by default and only allows approved destinations. Why this matters: Semantic models can inadvertently expose sensitive data when connecting across sources or workspaces. By enforcing policies at the connection level, organisations gain stronger control over data movement and reduce hidden security risks. Learn more.

• The new semantic model settings pane improves usability by allowing users to view and edit settings without leaving their workspace context. Opening as a side panel, it eliminates page navigation and keeps the current view intact. With grouped sections, collapsible layouts, and built-in search, managing settings becomes faster and more intuitive. Why this matters: This change is important for productivity, particularly for teams managing multiple models. Learn more.

Microsoft Fabric

• The traditional ETL model is no longer sufficient for modern data needs, as organisations increasingly require pipelines that interact with business processes and real-time decisions. Fabric introduces pipelines that can handle long-running tasks, external integrations, and approval-based workflows. Why this matters: This approach improves governance and operational efficiency by consolidating orchestration into one place, making it easier to manage complex, cross-functional data processes in enterprise environments. Learn more.

• OneLake catalog is now embedded directly within Microsoft Foundry, enabling users to discover and select governed data without switching tools. This integration simplifies how teams connect data to AI workflows by allowing them to browse assets, review metadata such as ownership and sensitivity, and immediately use them as knowledge sources. Why this matters: It removes friction between data discovery and AI development, accelerating the creation of grounded AI solutions. Learn more.

• Data Agents in Fabric now support service principal authentication, allowing applications to interact with agents using their own secure identity rather than relying on user credentials. This enables automation, background processing, and integration with enterprise systems. Why this matters: It improves security, scalability, and manageability, making it easier to move data agents from prototypes into production-ready solutions in real-world enterprise scenarios. Learn more.

• By supporting Eventhouse user-defined functions and materialised views, and shortcut tables, Fabric Data Agents can now tap into richer, pre-built logic and high-performance datasets. This means users can ask questions without needing to understand underlying query structures. Why this matters: It bridges the gap between advanced data engineering constructs and business-friendly AI interfaces, enabling more accurate and efficient self-service analytics. Learn more.

• Fabric’s architecture brings together multiple analytics capabilities into one platform, creating both opportunities and complexity for solution design. The Well-Architected framework helps organisations address this by focusing on key principles such as performance efficiency and operational excellence. Why this matters: It is invaluable for enterprise environments where shared resources and large-scale data workloads require careful governance to avoid inefficiencies, control costs, and ensure consistent performance across teams. Learn more.

• A new storage reporting capability in OneLake provides granular insight into how storage is distributed across items within a workspace. Users can explore, sort, and refresh reports to identify storage-heavy assets. Why this matters: It removes the need for manual analysis, improves transparency, and supports more effective governance, especially as data volumes grow and storage costs become harder to manage. Learn more.

• Fabric introduced Incremental Liquid Clustering, a more efficient clustering approach that targets only newly added or poorly structured data during optimisation. By avoiding full-table rewrites, it reduces maintenance overhead and improves scalability. Why this matters: It enables faster processing and lower costs for large datasets, while still delivering the performance benefits of well-structured data layouts for analytics workloads. Learn more.

• OneLake storage tiers introduce hot, cool, and cold storage options, allowing organisations to align storage costs with data access patterns. Frequently used data remains in the hot tier, while less active data can be moved to lower-cost tiers. Why this matters: It enables significant cost savings without sacrificing accessibility, helping organisations manage growing data volumes more effectively while maintaining compliance and performance requirements. Learn more.

Microsoft 365 Copilot and Copilot Studio

• Microsoft 365 Copilot’s May 2026 update introduced major model upgrades, including GPT‑5.5 Instant and Claude Opus 4.8, alongside a redesigned app experience and enhanced Notebooks for organising work. New capabilities across Teams, Outlook, Word, and PowerPoint improve productivity through smarter summaries, call delegation and personalised writing. Why this matters: The updates significantly expand Copilot’s intelligence, model flexibility and real‑world usability, reinforcing its role as a core productivity layer in Microsoft 365. Learn more.

• Microsoft 365 Copilot has achieved ISO 42001 recertification for the second consecutive year, reinforcing its commitment to responsible AI governance. Over the past year, the platform has expanded to a multi-model approach and strengthened risk management and oversight processes. Why this matters: It demonstrates how enterprise AI can scale while maintaining transparency, compliance, and user trust, which are critical factors for wider adoption across regulated industries. Learn more.

• New Copilot Cowork capabilities expand how teams interact with AI by incorporating organisational knowledge through plugins and connectors. This allows users to bring together data from multiple systems and execute tasks within a single workflow. Why this matters: It reduces fragmentation between tools and introduces a more unified, action-oriented approach to productivity powered by AI. Learn more.

• Microsoft has added Mistral Medium 3.5 to Copilot Studio, expanding model choice for building AI agents with stronger multilingual performance and in-region data processing options. Administrators maintain control through opt-in governance and configuration settings. Why this matters: It reflects a shift toward multi-model flexibility, allowing organisations to select the most appropriate AI model while maintaining compliance and control over data handling. Learn more.

• Copilot Studio introduces stronger governance controls, improved workflow capabilities, and deeper integration with business applications. Enhanced visibility into agent performance and security helps organisations manage risk, while intelligent workflows support more advanced automation scenarios. Why this matters: As AI agents scale, organisations need robust oversight and predictable behaviour, ensuring automation remains controlled, auditable, and aligned with enterprise policies. Learn more.

• Real-time voice agents in Copilot Studio are positioned as customer-facing AI systems that need more than good conversation design. The guidance focuses on governance foundations, security, compliance, and production readiness so teams can scale voice experiences responsibly. Why this matters: Natural, context-aware conversations raise expectations, but success depends on whether organisations can run them reliably, integrate them with business systems, and keep control as usage grows. Learn more.

Microsoft Foundry

• Microsoft Foundry’s model router now has an open-source evaluation pipeline that measures quality, cost, and latency in one run, helping developers judge whether automatic routing beats a fixed model choice. It also shows model distribution, value metrics, and optional hand-off into Foundry’s enterprise evaluation tooling. Why this matters: Teams can test routing decisions with their own prompts before committing to production architecture and spend. Learn more.

• Microsoft Foundry Agent Lab introduces nine open-source demos that build agent capabilities step by step, from a minimal prompt-based agent through tool calling, web search, code execution, retrieval, MCP integration, toolbox governance, and self-hosting. All demos use the same SDK and model router deployment. Why this matters: Developers can learn one concept at a time, which makes production-grade agent design easier to understand, reuse, and scale. Learn more.

• Microsoft Foundry has added Anthropic’s Claude Opus 4.8 as a model choice for developers building sophisticated applications. Its strengths span software development, long-form reasoning, automation, and regulated document workflows such as legal, finance, and cybersecurity use cases. Why this matters: Teams can now evaluate a highly capable model inside the same platform they already use for model comparison, deployment, and enterprise controls. Learn more.

• Foundry Local 1.1 adds live transcription, text embeddings, and Responses API support, extending local AI development beyond chat into speech, semantic search, and structured agent interactions. Microsoft also reduces package size, broadens .NET compatibility, and separates WebGPU support into an optional plugin. Why this matters: Developers can build richer on-device AI experiences with lower latency, better privacy, and fewer cloud dependencies while keeping portability across multiple languages and platforms. Learn more.

Databricks

• Enterprise leaders are moving AI agents into core workflows such as HR, finance, fraud detection, and creative operations, but they are doing so with governance built in from the start. Databricks highlights five practices including formal risk reviews, stronger monitoring, and clearer ROI discipline. Why this matters: Agentic AI only creates lasting value when speed, trust, and cost control are managed together rather than treated as separate workstreams. Learn more.

• Unity Catalog now offers broader Apache Iceberg support, including managed and foreign tables, Iceberg v3 capabilities such as deletion vectors and row tracking, and sharing to Iceberg-compatible clients. Databricks argues that the catalogue is becoming the defining layer of the open lakehouse. Why this matters: Enterprises increasingly need one governance and interoperability model across engines, formats, and AI workloads rather than fragmented platform-specific controls. Learn more.

• Lakebase Change Data Feed removes much of the manual work involved in extracting change data from operational databases by exposing every table’s changes once through Unity Catalog managed tables. Downstream tools, models, and agents can then read the same governed feed directly. Why this matters: Operational data can now act as a native Bronze layer without connector sprawl, making medallion-style architectures and agent-first development far easier to maintain. Learn more.

• Telecommunications firms are investing heavily in AI, yet many initiatives struggle to reach production scale because “data debt” gets in the way. Databricks highlights a unified semantic layer in Unity Catalog as the answer, supported by governance tools such as attribute-based access control and masking. Why this matters: Regulated industries need AI systems that understand their domain language and respect compliance boundaries from end to end. Learn more.

• Databricks Genie aims to solve the “last mile” of data democratization in financial services by letting business users ask plain-English questions against governed lakehouse data and receive auditable answers quickly. Databricks argues that many firms already have strong technical data platforms, but insight remains bottlenecked by specialist teams. Why this matters: Decision-makers need compliant self-service access if data investments are to influence day-to-day business choices more directly. Learn more.

• Databricks frames incident prevention as a data access problem rather than an on-call problem. Metrics, logs, traces, and SLO burn rates already exist, but teams cannot query them quickly enough to guide decisions before something breaks. Why this matters: Every avoidable incident affects roadmap delivery, customer trust, and support costs, making proactive observability a business issue as much as a technical one. Learn more.

• Databricks is addressing a practical risk in agentic AI by adding governance for MCP tools through Unity Catalog and Unity AI Gateway. Administrators can now control which tools an agent may call, apply conditions to tool use, and capture a full audit trail of requests and payloads. Why this matters: Destructive agent actions are no longer theoretical, and organisations need both prevention and forensic visibility. Learn more.

Open-Source

• Perplexity has open-sourced Bumblebee, a read-only endpoint scanner that checks developer machines for risky packages, browser extensions, and AI tool configurations during software supply-chain incidents. It fits into a broader workflow where threat signals become reviewed catalogue updates and then machine scans. Why this matters: Protecting end users increasingly starts with securing developer environments, not just production systems, and Bumblebee gives security teams a faster way to assess exposure. Learn more.

• Meta’s SAM 3 is a unified foundation model for detecting, segmenting, and tracking objects in images and videos using either text or visual prompts. The major step forward is promptable concept segmentation, which can exhaustively find all instances of an open-vocabulary concept rather than just segmenting one prompted object. Why this matters: Vision systems become far more useful when they can understand rich concepts across both images and video without fixed label sets. Learn more.

• DuckDB’s new Quack protocol lets separate DuckDB instances communicate over HTTP, turning DuckDB into a client-server database with multiple concurrent writers. It keeps setup simple, uses DuckDB’s own serialisation path, and is designed to handle both small transactions and large analytical transfers efficiently. Why this matters: Quack addresses one of DuckDB’s most important historical limitations, making it more practical for shared operational and multi-user workloads without abandoning its lightweight ethos. Learn more.

• Schema changes rarely fail in just one way, and Polars makes that explicit by separating new columns, missing columns, widened types, and genuinely breaking semantic changes. It then links each case to the right read or write parameter by storage format. Why this matters: Pipeline resilience depends on correctly diagnosing the type of schema drift first, rather than applying blunt fixes that hide deeper problems. Learn more.

• Apache Iceberg 1.11.0 is a substantial release, with more than 1,000 commits from over 200 contributors, and its biggest step forward is a more complete REST Catalog protocol. Remote scan planning now lets catalog servers plan scans and return only the relevant file tasks, while freshness-aware loading, idempotency keys, and view registration improve efficiency and reliability. Why this matters: It makes Iceberg easier to operate at scale across engines and streaming workloads. Learn more.

Industry

• Anthropic argues that Zero Trust needs to be adapted for agentic systems because AI agents do more than answer prompts. They interpret goals, choose tools, persist context, and execute multi-step actions. The framework focuses on cryptographically rooted identities, task-scoped permissions, protected memory, and defensive operations that assume breach from day one. Why this matters: Traditional access controls are not enough when agents can misuse legitimate permissions at machine speed. Learn more.

• Google Cloud’s “How Google Does It” collection pulls together 15 security explainers that show how Google approaches core cybersecurity challenges at scale, from threat detection and red teaming to SRE for security, cloud forensics, and AI agents for defenders. Why this matters: The series translates Google’s internal security practices into concrete lessons that enterprise teams can use to modernise operations and strengthen resilience. Learn more.

• OpenAI sets out a practical framework for independent evaluation of frontier models, arguing that useful assessments must describe both the claim being tested and the evidence that the result is valid. It highlights three evaluation goals such as capability elicitation, safeguard performance, and model comparison, while warning about reward hacking, refusals, contamination, and broken tasks. Why this matters: Credible third-party evaluations are becoming essential to safety and trust for more capable AI systems. Learn more.

• Perplexity has launched a finance-specific version of Computer aimed at professional research, analysis, and decision workflows. Teams can connect licensed data sources through MCP connectors or use built-in finance tools, then turn that data into artefacts such as tearsheets, dashboards, memos, and charts. Why this matters: Finance work depends on accuracy and traceability, and Perplexity is trying to connect cited data directly to finished, auditable outputs. Learn more.

• Gartner highlights six trends shaping software engineering technology adoption in 2026, with productivity and developer experience remaining the top value driver. At the same time, only 35% of software engineering leaders report significant ROI from AI in the software development life cycle. Why this matters: Leaders are under pressure to invest in AI tools and platforms, but success depends on linking adoption to business outcomes rather than stopping at productivity claims. Learn more.

May’s developments reinforced a clear message: the next phase of analytics and AI will be defined less by what the technology can do, and more by how well organisations can govern, integrate, and operationalise it. Across platforms, the emphasis shifted towards trusted agent workflows, stronger observability, reusable patterns, and more open interoperability. Whether teams are extending Fabric, scaling Copilot, evaluating Foundry, adopting Databricks capabilities, or tracking open-source innovation, the priority is now turning promising tools into secure, sustainable, enterprise-ready capabilities that can deliver value beyond the pilot stage.

Stay in the Know

Get notified when we post something new by following us on LinkedIn and X.