Microsoft 365 Copilot is transitioning to a multi-model architecture that allows organizations to choose between various AI models for specific tasks. While Microsoft-hosted models remain the default, administrators can now enable third-party options like Anthropic’s Claude or models from xAI. This flexibility ensures that users can match the specific strengths of different models, such as deep reasoning or creative drafting, to their unique workflows.

Source

Microsoft is launching a webinar series focused on unifying identity, access, and data protection to address the unique risks posed by artificial intelligence. The initiative aims to help organizations manage sensitive data as it flows through AI prompts, browsers, and non-human actors. By integrating Microsoft Entra and Microsoft Purview, the company intends to provide a more cohesive strategy than traditional legacy security models.

Source

AI coding agents often ignore updated documentation because they form a task execution plan before retrieving external data. This behavior stems from the model relying on statistical defaults learned during its initial training phase. When a user provides a prompt, the agent immediately constructs a strategy based on its internal knowledge base.

Source

The Trump administration has requested that OpenAI delay the broad release of its upcoming GPT-5.6 model due to potential national security concerns. In response, CEO Sam Altman announced that the model will initially launch only in a limited preview capacity for a select group of enterprise customers. This shift marks a significant intervention by the federal government into the deployment timelines of advanced artificial intelligence systems.

Source

AI-assisted development allows teams to build internal applications at unprecedented speeds, but this efficiency often bypasses traditional security oversight. Many organizations are unknowingly integrating open source libraries and frameworks that lack long-term support or active maintenance. This trend is rapidly increasing technical debt and creating hidden security risks that only surface during major compliance audits or modernization efforts.

Source

Google is restructuring its internal AI coding "strike team" within the DeepMind division only two months after its initial formation. This strategic shift aims to enhance the programming proficiency of the Gemini model to better compete with rivals like Anthropic. The reorganization involves expanding the team's responsibilities to integrate advanced development techniques into the model's core training.

Source

OpenAI has reportedly postponed its initial public offering until 2027 to maintain a target valuation of one trillion dollars. CEO Sam Altman rejected the alternative of an earlier public launch at a lower valuation, according to internal reports. This decision follows a confidential application for a U.S. IPO that aimed for record-breaking market figures.

Source

Anthropic has introduced Claude tag, a feature that integrates its AI directly into Slack as a persistent, collaborative team member. Unlike traditional bots that require manual prompts, this tool operates in an "ambient mode" to ingest channel conversations and internal documentation. By building a comprehensive graph of a company’s operations, Claude can proactively execute workflows and generate code without users leaving their primary communication interface.

Source

Anthropic is currently testing mobile support for Claude Cowork to allow users to manage complex tasks from their smartphones. Claude Cowork is a desktop-oriented agentic mode designed to handle long-running knowledge work and background processes. While the existing Claude Code tool focuses on development, Cowork specializes in generating documents, managing spreadsheets, and analyzing local file systems.

Source

OpenAI has announced a staggered rollout for its upcoming GPT-5.6 model, prioritizing a small group of partners over a general public release. This shift follows a request from the federal government to review frontier artificial intelligence models before they are deployed at scale. The move signals a tightening relationship between major AI developers and Washington regarding national security and safety standards.

Source

Microsoft Threat Intelligence has uncovered a multi-stage cyberattack campaign targeting the hospitality industry in Europe and Asia since April 2026. The threat actors use "authentication laundering" by abusing legitimate services like Calendly and Google redirects to bypass email security filters. Phishing lures often involve guest complaints or room inquiries to trick staff into downloading malicious ZIP archives containing fake image shortcuts.

Source

Microsoft has officially acknowledged that a bug within the Background Intelligent Transfer Service (BITS) is responsible for slow shutdown times in Windows 11. This service, which manages background downloads and updates, was failing to stop promptly when users attempted to turn off or restart their computers. The issue frequently caused the "Shutting down" screen to persist for several seconds longer than necessary while the system waited for BITS to terminate.

Source

Chinese cybersecurity firm Qihoo 360 has announced a new AI-driven vulnerability discovery system designed to compete with Anthropic’s Mythos model. CEO Zhou Hongyi described the American tool as a strategic asset and argued that China requires equivalent capabilities for national deterrence. Rather than relying on brute-force computing power, the company utilizes a specialized multi-agent swarm architecture to identify software flaws.

Source

Modelplane is a new open-source control plane designed to coordinate AI inference across diverse hardware environments. It addresses the complexity of managing GPU fleets distributed across public clouds, specialized neoclouds, and on-premises data centers. By automating these processes, the tool eliminates the need for organizations to build custom coordination layers manually.

Source

Lenovo has issued a warning that the current high costs of DRAM and NAND flash memory are unlikely to return to previous lows. The PC manufacturer suggests that these elevated prices could represent a permanent shift in the market, lasting well beyond 2030. This trend marks a significant departure from historical cycles where hardware costs typically decreased over time.

Source

Microsoft is automatically rolling out Windows 11 25H2 to all eligible unmanaged Home and Pro PCs, with no user approval required. The primary driver is the end of support for Windows 11 24H2 on October 13, 2026, after which those editions will no longer receive security updates. The rollout also prepares consumer devices for Windows 11 26H2, expected in fall 2026 on the same underlying platform. For devices already on 24H2, the upgrade is delivered as a small enablement package rather than a full OS reinstall. Managed enterprise devices are excluded from this automatic rollout.

Source

Microsoft Defender Antivirus provides a robust set of PowerShell cmdlets for managing security features on Windows 10, Windows 11, and Windows Server 2016 or later. These commands allow for the configuration of next-generation protection features, including exploit protection and real-time monitoring. Administrators can use these tools to automate security tasks and ensure consistent protection across a fleet of devices.

Source

Apple is reportedly planning to disrupt its traditional silicon release cycle to accelerate the development of high-performance processors. The company intends to launch the standard M6 chip this fall but will likely skip the Pro and Max variants of that generation. This strategic shift aims to bring the M7 architecture to market up to six months earlier than originally scheduled.

Source

Anthropic's Claude is experiencing a significant surge in paid consumer subscriptions, challenging the market dominance of OpenAI's ChatGPT. Transaction data from millions of U.S. consumers indicates that Claude's revenue and paying user base grew by approximately 75% in early 2026. While the company was previously viewed as a developer-centric niche, it is now successfully attracting a broader audience of individual subscribers.

Source