photo by Microsoft

Microsoft has officially rolled out its June 2026 update for Microsoft 365 Copilot, introducing a wave of automated “agentic” workflows, advanced multi-model capabilities, and tighter administrative controls designed to boost workplace productivity.

The era of action. Copilot Cowork reached general availability

The headline of this month’s release is the general availability of Copilot Cowork, a fully agentic system powered by Work IQ. Unlike traditional AI assistants that generate drafts or recommendations, Cowork takes user-defined tasks and executes them from start to finish to deliver a completed final product.

Available via usage-based billing, Cowork introduces several major enhancements:

• Multi-model flexibility: the system automatically pairs tasks with the best-suited AI model. It utilizes OpenAI’s GPT 5.5 Thinking mode for deep research and selects Anthropic Claude models for visual-heavy tasks like PowerPoint generation.
• Expanded integration: a new wave of enterprise plugins is now supported, including Miro, monday.com, Harvey, and the full Dynamics 365 portfolio.
• Custom skills & visuals: Users can build and save personalized “skills” to standardize recurring workflows. Furthermore, powered by “ChatGPT Images 2.0,” Cowork can now directly create and edit graphics, illustrations, and images using an organization’s branded asset templates.
• Cross-platform access: Cowork now supports mobile functionality on iOS and Android, works via the Edge browser across intranet sites, and sends push notifications for long-running tasks.

Smarter chat, deep citations, and Copilot Notebooks upgrades

Microsoft 365 Copilot Chat and the core Copilot app have received significant analytical upgrades through Work IQ. Copilot can now reason over structured enterprise data, pulling answers directly from Power BI reports and Dataverse business records using natural language queries.

To increase transparency, the new “deep citations” feature allows users to click links that lead directly to the specific section of a referenced Word or PowerPoint file. Additionally, a new “Regenerate” action lets users quickly swap models or retry a prompt in place.

Looking ahead, a new “Vision” feature is set to roll out in July, allowing users to show Copilot live screens or camera feeds during voice conversations to troubleshoot errors or explain complex dashboards. Meanwhile, Copilot Notebooks has expanded its access to standard Chat users and will allow Outlook emails to be added as references starting in July.

App-specific enhancements. Office gets more flexible

Microsoft has brought targeted updates to its core productivity suite to streamline editing and formatting:

• Word: users can now select Anthropic models for document editing, utilize a “Catchup” card to see recent document changes, insert AI-generated images inline, and apply automated edits based on user comments. These capabilities have also expanded to iOS devices.
• Excel: introduces “Personalization” for standing user preferences (such as formula and layout styles) and “.Rules” sheets to lock in workbook-specific formatting guidelines for the whole team.
• PowerPoint & Outlook: PowerPoint gains a “Brand Kit Picker” to automatically apply approved corporate layouts, alongside the ability to reference OneDrive and SharePoint libraries. Outlook web users can now refine email drafts directly in the compose canvas, while Classic Outlook for Windows adds direct access to Copilot settings this July.
• Copilot Agents: the newly available “Planner Agent” helps users manage, prioritize, and build structured plans via natural language.

New tools for IT admins

To support these resource-intensive AI features, Microsoft has added robust management tools within the Microsoft 365 admin center.

A new Cost Management Dashboard allows administrators to track usage-based billing credits, enforce budgets, and set hard caps on tools like Copilot Cowork. Starting in July, managers will get deeper team-level credit reporting through Insights, alongside public preview metrics for custom agent adoption. Admins can also now publish custom “organization prompts” directly into the user Prompt Gallery and leverage controls to restrict or permit the upcoming “Vision” feature.

Finally, security is heavily reinforced via Microsoft Purview. Data Loss Prevention can now block external emails from being used as grounding data or summaries in Copilot. Purview controls have also been fully extended to Cowork, ensuring that sensitivity labels, audit logging, and eDiscovery apply seamlessly to all agentic interactions.

Claude agents on Azure now powered by NVIDIA Blackwell Ultra

photo by Nvidia

Anthropic’s Claude models in Microsoft Foundry – hosted on Microsoft Azure and running on NVIDIA GB300 Blackwell Ultra GPUs – are now generally available, giving Azure-native enterprises a powerful new way to build autonomous and domain-specific AI agents.

As agentic AI continues to drive enterprise innovation and becomes more autonomous, organizations need access to computing power to build and deploy specialized agents to accelerate essential business tasks. And having great inference performance and efficiency reduces total cost of ownership and drives positive company results.

High-performance AI infrastructure with NVIDIA

With Claude AI in Microsoft Foundry running on NVIDIA GB300 NVL72 systems with NVIDIA Quantum-X800 InfiniBand networking, enterprises can now build and run more powerful agentic systems, including autonomous and specialized sub-agents that can work across business domains to perform advanced tasks.

NVIDIA is working with Anthropic to extend developer capabilities by integrating NVIDIA tools into the Anthropic stack. “Through NVIDIA verified agent skills, enabled by access to NVIDIA accelerated computing, enterprises can embed AI agents deeply into their business and use them as the operating system for the organization” – said NVIDIA.

To ensure proper governance, enterprises can run Claude agents on Azure by using the NVIDIA Secure Agent Workspace Reference Design. “It provides a blueprint for running autonomous agents in a governed environment where identity, network access, credentials and runtime policy are controlled at the infrastructure level” – the company stated. This release builds on the strategic partnership Microsoft, NVIDIA, and Anthropic announced in November 2025.

Consistent billing via Claude Consumption Units

The Redmond Giant has also clarified how Claude models are billed in Microsoft Foundry using a construct called the Claude Consumption Unit (CCU): “A Claude Consumption Unit (CCU) is a unit of measure used solely for invoicing of Claude models. CCU billing applies to all Claude models offered in Foundry.”

This format change does not represent a price increase, as costs remain driven by Anthropic’s published per-model token rates. It requires no changes to application SDKs or APIs. Furthermore, CCU spend fully decrements the customer’s Microsoft Azure Consumption Commitment (MACC) just like any other Azure Marketplace consumption. Detailed per-model token and request metrics continue to be visible in the Monitoring tab of the Microsoft Foundry portal.

Excel gets a major AI upgrade with new tools for finance experts

photo by Microsoft

The trajectory of new technology reaching the world follows a familiar pattern: wave after wave, from the PC to the relational database to the cloud, tools reach developers first, and finance is often next. For finance professionals, a better tool provides an edge, and the job requires modeling reality a little more precisely than yesterday. For decades, that tool has been Excel – where the quarter gets closed, the forecast gets argued line by line, and every number traces back to a source.

When AI enters finance, it must clear that same high bar: showing its work, using trusted data, and tracing every calculation. While many AI tools claim to be built for finance, Microsoft 365 Copilot in Excel is proving it in practice.

Across Financial Planning and Analysis (FP&A), Accounting, Tax, Compliance, and Treasury, Microsoft Finance runs Copilot in Excel in real workflows. This integration allows teams to spend less time hunting for information or rebuilding analyses, freeing them to apply judgment to critical decisions. The finance team shapes the tool as much as they use it, identifying shortcomings and pushing the product toward their demanding standards. Now, Microsoft is introducing new features built specifically for financial professionals, featuring repeatable workflow skills, new financial connectors for trusted data, and improved traceability.

Built for the complexity of finance work

Before shipping new Copilot capabilities in Excel, Microsoft evaluates them across graded levels of task complexity and benchmarks that reflect daily finance operations. This ensures the AI can deliver multi-step workflows with trusted, verifiable results rather than just completing single tasks.

Meeting this high bar involved partnering with the Financial Modeling Institute (FMI), the global body that credentials the industry’s most demanding modelers. FMI’s library of real-world financial modeling cases has become a foundational component of how Copilot in Excel is evaluated for finance work.

Tuned to professional standards with repeatable skills

Microsoft is introducing skills that allow teams to define exactly how Copilot should complete common processes, such as building a DCF, closing the books, refreshing a monthly reporting model, or preparing a variance analysis. Instead of starting from scratch each time, a skill guides Copilot through the necessary steps, applying the correct structure and formatting to produce an output that is easy to review, reuse, and trust.

A library of sample finance skills is currently available, and users can build custom skills using an open-standard markdown file. By saving a “SKILL.md” file in OneDrive, Copilot can automatically pick it up to construct a three-statement model or a board package based on the user’s defined process.

Furthermore, developers and partners will soon be able to build and deploy skills through the Microsoft Marketplace and Microsoft 365 Admin Center. Microsoft has already begun working with an initial wave of partners, including finance and ERP providers like LSEG, Ramp, Rogo, Samaya AI, Velixo, and Vena. 

Copilot also adapts to individual workflows. With Personalization, users set preferences once for consistent application, while workbook rules capture structural, naming, and formula conventions within a sheet that follows the specific file.

Grounded in trusted financial data connectors

Copilot in Excel now connects directly to the data financial professionals rely on, bringing market data, fundamentals, and research directly into the workbook to eliminate manual data pulls. Expanding upon the LSEG and Moody’s connectors released in May, Microsoft is adding more financial data connectors to pull public and private market data:

• CB Insights: Brings predictive intelligence on private companies and markets into Excel to support sourcing companies, evaluating emerging markets, and M&A workflows.
• Daloopa: Provides audit-ready fundamentals sourced from SEC filings, investor presentations, and public materials to update operating models and reduce manual data entry.
• FactSet: Connects workflows to institutional financial and alternative data for modeling, screening, and market analysis (currently in preview, generally available in July).
• Morningstar: Supplies investment research, ratings, and portfolio analytics to evaluate holdings and support asset allocation decisions.
• PitchBook: Delivers institutional-grade private capital market intelligence, deal histories, and fund data directly into Excel for diligence and investment screening.
• S&P Global – Deterministic Retrieval: Developed by Kensho, this provides structured, API-driven access to S&P Global data for predictable, cited results and full orchestration control.

Note: Third-party connectors and data providers may require separate licensing or subscriptions from the respective provider.

Unlocking key finance workflows

Combined with Work IQ for grounding in work context, these features unlock specific, everyday financial scenarios using designated skills and connectors:

• Close the books: Compare last quarter’s actuals to plan using internal forecasts and planning decks. Identify the five largest variances across revenue, expenses, margins, and cash flow, explain drivers, and draft an executive summary using “@variance-analysis”.
• Update the forecast: Roll forward current forecasts using the latest team assumptions and budgets, incorporate market data, reconcile changes against the operating plan, and summarize key drivers using “@model-update”.
• Build the valuation model: Utilize “@comps-analysis” to build a DCF, comparable company analysis, and sensitivity model by pulling fundamentals, analyst expectations, and transaction multiples.
• Find the next acquisition: Use “@deal-screening” to identify and rank acquisition candidates matching internal strategy criteria by combining company performance, market signals, and funding history.
• Analyze portfolio performance: Assess a portfolio against investment objectives and committee materials using “@portfolio-monitoring” to identify concentration risks and pull fund analytics.
• Stay ahead of earnings: Use “@catalyst-calendar” to analyze earnings results, estimate revisions, and consensus forecasts to identify sentiment shifts and track critical investor developments.

Controllable and traceable by design

In finance, an answer alone is insufficient; professionals must know exactly how it was derived. Finance teams require visibility into changes, confidence in the methodology, and a clear audit trail.

To address this, users can now choose to “Plan with Copilot” before taking action. This feature outlines which ranges, worksheets, formulas, and assumptions the AI intends to update, while surfacing clarifying questions. Once changes are executed, every edit remains traceable with links back to the affected cells. Furthermore, changes are now explicitly attributed to Copilot alongside human collaborators within the “Show Changes” pane, allowing the AI to function like a trusted analyst that proposes a path, explains its approach, and ensures total transparency.

Personalization, workbook rules, pre-built skills, federated Copilot connectors, “Plan with Copilot”, and Copilot attribution in “Show Changes” are generally available for Microsoft 365 Copilot customers across Excel for Web, Windows, and Mac.

Custom skills are already available via the Insiders channel for Windows and Mac, and will roll out to general availability across Excel for Web, Windows, and Mac next month. Partner-built skills are scheduled to arrive in Q3 2026.

Microsoft Teams tightens security. Bots won’t breach your meetings

photo by Microsoft

AI note-taking has become a familiar part of the meeting experience, helping people keep track of important details while staying present. However, as AI meeting tools have become more common, new challenges have emerged. Bots have begun joining meetings that participants never intended them to attend. For example, after connecting a third-party service to a meeting, some users have found that its bot continues joining future meetings automatically.

Unexpected participants can create security and privacy risks, particularly when sensitive information is being discussed. Organizations need confidence that the right people and tools are participating in their discussions. That is why a new Teams admin policy has been introduced, designed to give organizations more visibility and control over external bots in their meetings. This new experience helps organizers identify bots and adds safeguards before they are admitted, giving organizations greater confidence that only the intended participants and tools will be present.

A new admin policy for managing external bots

The new policy in the Teams Admin Center, Manage external bots and their access to meetings, can be assigned to individual users or specific groups. Admins can choose between two settings. The default option is When detected, require approval before joining, which ensures Teams detects bots, puts them in the lobby, and requires explicit organizer confirmation before they are admitted. Alternatively, admins can choose Do not detect bots to disable the experience entirely.

When enabled, Teams automatically detects potential bots, places them in the meeting lobby, clearly identifies them, and prompts organizers to confirm admission. Even in meetings where organizers allow participants to bypass the lobby, bots identified through this policy will continue to require approval before joining. As a useful tip, organizers should set the meeting option Who can admit from lobby to organizers and co-organizers only to ensure no unintended participant can admit unwanted participants or bots from the meeting lobby.

Detecting AI bots more intelligently

First, Teams’ ability to distinguish between bots and human participants as they join a meeting has been strengthened. Teams now uses a combination of behavioral and infrastructure signals to identify bots with a higher degree of accuracy.

Alongside these improvements, a registration path will soon be introduced for independent software vendors (ISVs) that build meeting experiences for Microsoft Teams. Through the Teams Bot Identification Program, bot providers will be able to register with Microsoft and include a self-identification marker in their join requests. When Teams recognizes that marker, it can identify the bot as a known participant.

Currently, Microsoft is working with a limited set of ISVs to preview this capability and validate the experience before broader availability, with additional details to be shared in the future.

Giving organizers clearer visibility and safeguards

When bots are detected, they are directed to the meeting lobby and visually distinguished from other participants. This makes it easier for organizers to see who is waiting in the meeting lobby to join and make informed admission decisions. Instead of scanning a long list of names, organizers can quickly assess who is waiting to join and identify potential risks at a glance. Participants in the lobby are now grouped into two distinct categories: Waiting for verified for standard participants and registered bots, and Suspected threats for unregistered or system-identified bots.

Safeguards designed to reduce the accidental admission of these identified bots from the meeting lobby have also been added. There is no longer a one-click Admit option for identified bots. Instead, the system displays confirmation prompts when admitting participants that include bots, as well as clear warnings when organizers choose Admit all and bots are included in the queue. Admitting a bot should be a deliberate decision, not something that happens by mistake.

Rolling out this new experience and looking ahead

As this capability rolls out in June 2026, the existing CAPTCHA verification experience will begin to be retired and replaced with this more comprehensive approach to managing external bots in Teams meetings.

This is just the beginning, and the experience will continue evolving based on customer feedback. Additional admin controls and monitors are already being explored, including allow lists for approved bots, organization-wide policies to block external bots entirely, admin reports and audit logs on the detection and presence of bots, and more granular controls aligned to different security requirements. As these experiences evolve, transparency and control remain essential to helping organizations embrace new meeting innovations with confidence.

OpenAI’s flagship GPT-5.6 Sol takes the lead over Claude Mythos 5

photo by OpenAI

OpenAI has announced a limited preview of its new GPT-5.6 model series, featuring the flagship Sol, the balanced Terra, and the fast, low-cost Luna. Due to US government restrictions, initial access is limited to a small group of trusted partners, following a capability review with officials. The models deliver massive improvements in agentic performance, coding, and cybersecurity. GPT-5.6 also introduces layered safety upgrades and deep reasoning modes. OpenAI plans to make the entire series broadly available in ChatGPT, Codex, and the API in the coming weeks.

OpenAI introduces the GPT-5.6 series: Sol, Terra, and Luna

OpenAI has officially announced the launch of a limited preview for its next-generation artificial intelligence lineup, designated as GPT-5.6. This new model family consists of three distinct tiers tailored to different performance requirements and budgets. In its official blog post, the company stated: “We’re beginning a limited preview of the GPT‑5.6 series: Sol, our flagship model; Terra, a balanced model for everyday work; and Luna, a fast and affordable model.”

In a shift from previous release strategies, the rollout of GPT-5.6 is subject to specific administrative constraints. Due to requests from the U.S. government, OpenAI is limiting the initial wave of access to a select group of trusted partners. The company previewed its deployment plans and model capabilities to government officials ahead of the announcement. However, OpenAI explicitly noted that this type of government-vetted access process should not become the long-term standard for the industry, as it delays the deployment of defensive tools to developers, enterprises, and cybersecurity professionals. The current phased approach is intended as a temporary step to secure a path toward broader availability.

Enhanced capabilities and benchmark milestones

GPT-5.6 Sol stands as OpenAI’s most powerful model to date, showing significant advancements in agentic operations across software engineering, biology, and security research. On the Terminal-Bench 2.1 evaluation, which measures command-line workflows involving multi-step planning and tool execution, Sol established a new state-of-the-art record score of 91.9%, surpassing Anthropic’s Claude Mythos 5. In quantitative biology, Sol outperforms GPT-5.5 on the GeneBench v1 benchmark while executing complex genomics analyses with higher token efficiency.

The release also introduces a new “max” reasoning effort setting to facilitate deeper processing times for complex queries. Furthermore, an “ultra” mode makes its debut, leveraging cooperative subagents to accelerate workloads that extend beyond the capacity of a single standalone model.

To counteract potential risks associated with these advanced capabilities, OpenAI implemented its most robust safeguard infrastructure to date. The safety architecture is configured to intercept and refuse malicious cyber requests and persistent misuse while ensuring that legitimate defensive activities, such as debugging, code review, and vulnerability patching, remain unaffected. As stated in the announcement: “GPT‑5.6 Sol is better at helping people find and fix vulnerabilities than reliably carrying out end-to-end attacks.”

The safety mechanisms operate across multiple layers, combining model-level refusal training with real-time misuse classifiers that inspect outputs during generation. High-risk instances trigger an automated pause, routing the conversation context to a larger reasoning model for safety verification before any text reaches the user. Persistent malicious patterns are tracked via account-level monitoring. To harden these defenses against adaptive adversarial tactics and universal jailbreaks, OpenAI dedicated over 700,000 A100-equivalent GPU hours to automated model-driven red-teaming, supplemented by ongoing expert human testing.

Pricing, infrastructure, and availability timelines

The naming convention introduced with GPT-5.6 decouples the model generation tier (5.6) from the capability and pricing segments (Sol, Terra, Luna). Pricing per million tokens has been structured as follows:

• Sol: $5 input / $30 output
• Terra: $2.50 input / $15 output
• Luna: $1 input / $6 output

The series introduces predictable prompt caching features, including explicit cache breakpoints and a 30-minute minimum cache life. Cache writes are priced at 1.25x the standard uncached input rate, while cache reads retain a 90% discount relative to standard input costs. Additionally, OpenAI is partnering with Cerebras to launch GPT-5.6 Sol at processing speeds of up to 750 tokens per second this July for select enterprise customers. General availability across ChatGPT, Codex, and the developer API is scheduled to roll out progressively over the coming weeks.

AI model choice in Microsoft 365 Copilot. Greater flexibility and control for enterprises

photo by Microsoft

In the face of rapid AI development, Microsoft has decided to transform its flagship productivity platform. The Redmond giant announced that Microsoft 365 is moving away from being locked into a single algorithm toward a multi-model architecture. This shift aims to provide organizations with the freedom to match AI tools to specific tasks without compromising corporate security standards.

No single artificial intelligence model is perfect for every scenario. While some systems excel at deep analysis and structured reasoning, others are optimized for processing speed, text summarization, or creative drafting. Microsoft 365’s new approach allows IT admins, creators, and end users to intentionally select the technology that best suits their current needs.

Multi-model architecture and the role of AI subprocessors

While Microsoft-hosted solutions remain the default foundation for Copilot experiences, organizations have gained the ability to use external models. These can function in two ways. The first is implementing external systems as official AI subprocessors for Microsoft Online Services. A prime example of this is the integration of Anthropic models.

In this scenario, the software operates within the Redmond giant’s contractual framework. This means that Microsoft Product Terms and the Data Protection Addendum (DPA) apply. Admins retain full, centralized control over the availability of a given solution, and its usage is 100 percent aligned with corporate governance and enterprise data protection commitments. All subprocessors are also contractually required to meet strict security and privacy standards.

Independent AI models as an opt-in flexibility

The second path is integration with independent AI models hosted outside Microsoft-managed infrastructure. Currently, the platform offers connections to systems from external providers such as Anthropic and xAI.

Choosing this option, however, comes with a different set of rules. Data may be processed outside Microsoft’s secure infrastructure, and provider-specific privacy policies apply. Explicit admin approval is required before these models become available to users or makers. Microsoft also recommends that organizations carefully review regional regulations and data residency considerations before enabling these tools.

Centralized governance and freedom for makers

Managing these new features takes place at the level of the Microsoft 365 admin center or Power Platform admin center. Corporate IT administrators can block or authorize external language models and decide which ones will generate responses. This allows companies to balance innovation with their organizational risk tolerance.

Meanwhile, in the Microsoft Copilot Studio, makers building custom agents can now independently select the primary model powering the bot’s reasoning. They have the ability to switch between algorithms to compare performance, utilize external options, or fall back to Microsoft-hosted defaults. Preview and experimental versions are clearly labeled as not recommended for general production use.

New capabilities for end users

The introduced changes are directly visible in daily operations. The Researcher agent in Microsoft 365 Copilot, designed to gather and analyze data from both the web and work content, can now simultaneously utilize models like OpenAI’s GPT and Anthropic’s Claude.

In “Auto” mode, the system first generates a report using GPT and then applies a second reasoning pass with Claude. This dual process strengthens the document’s structure, improves completeness, prioritizes reputable sources, and ensures key statements are grounded in citations. For more advanced workflows, the “Model Council” mode runs the same query through multiple deep-reasoning research agents in parallel, highlighting where outputs align or diverge so users can combine insights with greater confidence.

This new approach is built on a shared responsibility framework. Microsoft provides secure defaults and transparent tools, admins decide on access, and users choose tools intentionally based on the task. Crucially, all existing Microsoft 365 security, compliance, and permission controls remain in full force. Prompts, responses, and Microsoft Graph grounding data are not used to train foundational large language models (LLMs).

Microsoft retires AI history search in Edge after user backlash

photo by Microsoft

Microsoft has officially decided to discontinue its AI-powered history search feature for the Edge browser, marking a sudden end to an initiative that promised to streamline how users manage their browsing data. The cancellation, noted in a recent update to the Microsoft 365 Roadmap, comes just a year after the feature began its phased rollout.

A closer look at the cancelled AI feature

Introduced in June 2025 with the release of Edge 138, the AI-powered search was designed to prioritize productivity. By leveraging an on-device AI model, the feature allowed users to search for past web pages using natural language, synonyms, or descriptive phrases, rather than relying on exact keywords. The system was also built to account for typos, effectively removing the precision hurdles often associated with browser history searches.

Microsoft had emphasized the privacy-centric nature of the tool, assuring users that all processing would occur locally. The company stated that no data would be sent to the cloud, and provided IT administrators with the EdgeHistoryAISearchEnabled policy, granting them full control to enable or disable the functionality within their organizations.

User backlash and skepticism

Despite the technical safeguards, the rollout was met with immediate resistance. Upon its debut many users expressed distrust toward Microsoft’s data handling, labeling the feature as “creepy.” Beyond privacy concerns, critics dismissed the tool as unnecessary “AI bloat,” arguing that it was an unwelcome addition to the browser’s interface.

The cancellation is particularly noteworthy given Microsoft’s recent strategic emphasis on the “agentic AI era.” While the tech giant previously indicated it would be more mindful regarding the integration of Copilot features within Windows 11 apps, observers noted that such promises often amounted to rebranding rather than a genuine removal of AI tools. However, the decision to axe the Edge history search appears to be a definitive retreat from a specific AI-driven utility, rather than just a cosmetic change.

Microsoft has not provided a detailed explanation for the cancellation, offering only a brief apology to customers for the inconvenience. In a final update to the feature’s roadmap entry, dated June 25, 2026, the company stated:

Enhanced search finds sites in your History even when you use a synonym, phrase, or typo. After this feature is turned on, sites you visit will be shown in enhanced history search results. An on-device model is trained using your data, which never leaves your device and is never sent to Microsoft. Admins can use the EdgeHistoryAISearchEnabled policy to disable this feature.

Updated June 25, 2026: We have decided not to move forward with this change at this time. We apologize for any inconvenience.

Tidal introduces new strict policy for AI-generated music

photo by Tidal

Tidal has announced a new AI policy aimed at protecting human artists as AI music generation tools continue to improve. While the platform will still accept AI-generated music, these tracks will now be held to a “higher standard” of content integrity.

Starting mid-July, Tidal will automatically identify and tag tracks flagged as 100% AI-generated with a special icon. The platform will immediately take down AI music that exploits an artist’s voice or likeness and block tracks associated with fraud or artificial streaming. Crucially, music that is 100% AI-generated will no longer be monetized. Highlighting its focus on human creators, Tidal stated:

We acknowledge the ongoing debate regarding whether certain AI-generated music (e.g. AI-generated music developed from fairly and properly licensed models) should be entitled to earn royalties. This debate will continue as the technology advances and rightsholders and AI music platforms develop licensing models. Tidal’s priority is ensuring royalties go to original works directly produced, written, and performed by people. We will therefore not knowingly attribute royalties to music we identify as wholly AI-generated.

The policy arrives as streaming services face a massive influx of automated content. For context, synthetic uploads make up about 44% of Deezer’s daily intake (around 75,000 tracks), yet an Ipsos study revealed that 97% of listeners cannot tell the difference between human and machine creations.

Tidal’s restrictive stance contrasts sharply with Spotify, where AI music remains allowed and monetized. Spotify’s CEO recently urged listeners to stop calling AI music “slop” and instead embrace its creative potential. The platform has even partnered with Universal Music Group to test “legal and controlled” generative AI tools that allow subscribers to remix songs.

.