Microsoft Word for Windows and Mac now supports “Copy Link to Location,” creating cloud links that open a document at a specific piece of text or object instead of its beginning. The feature targets long-document reviews, but availability may vary as Microsoft rolls it out.

Source

Windows device recovery now spans automatic boot fixes, full cloud rebuilds, point-in-time rollback, and temporary Cloud PCs for failed or unavailable hardware. The biggest practical change for IT teams is the combination of Intune policies, OneDrive, Autopilot, and Windows settings backup, which can restore a user’s familiar environment after a rebuild or device replacement.

Source

Microsoft’s security review of Microsoft Edge 150 finds that none of its eight new computer and user settings require enforcement. Despite Edge 150 receiving multiple stable-channel updates in July, Microsoft continues to recommend the Edge 139 security baseline for enterprise deployments.

Source

Grockbot hides code, tool calls, and model selection behind a familiar chat interface while giving every agent its own cloud computer. Matthew Berman shows how agents can use websites, coordinate with one another, control a local PC, run scheduled routines, and learn repeatable tasks from demonstrations.

Source

OpenAI’s new Computer History feature gives ChatGPT a searchable timeline of activity across Mac apps and websites, letting users recover context from earlier work without taking continuous screenshots. It is opt-in and available to Pro, Business, and Enterprise subscribers, but its ability to summarize workplace activity creates an important privacy consideration for organizations.

Source

An Akira ransomware affiliate rebooted a victim into Safe Mode to disable Huntress and Microsoft Defender, but the restricted environment also caused Akira’s encryptor to fail. The attackers still stole credentials and data after entering through an MFA-less SonicWall VPN; earlier coverage examined how DeadLock suppresses Windows defenses and complicates recovery in a related ransomware campaign.

Source

Nvidia says its new NeMo Switchyard router can reduce AI agent evaluation costs by 74 percent by sending routine requests to cheaper models, although testing showed a six-point accuracy decline. This follows earlier coverage of Meta’s Switchboard model-routing project, but Nvidia is now offering its approach publicly alongside the Nemotron 3.5 Lightning open-weight model.

Source

A recent TONTOU attack showed that Spectre v2 defenses on AMD and Intel CPUs could still be bypassed; now researchers have demonstrated major Spectre vulnerabilities on commercial RISC-V processors. The attacks can extract arbitrary Linux kernel memory from a T-Head Xuantie C910 at about 338 bytes per second, overturning the assumption that RISC-V is largely immune.

Source

Cisco says Anthropic’s Mythos bug-finding model is triggering an urgent wave of replacements for networking equipment past its last day of support. The company calls this the “Mythos Effect,” as customers move security and networking budgets to replace legacy routers, switches, and other devices before automated vulnerability discovery exposes them.

Source

Microsoft keeps Microsoft 365, Entra ID, and Azure services available, but does not promise to restore a customer’s data to a specific known-good point after an attack. That makes independent, immutable backup increasingly important as stolen identities let attackers reach mailboxes, files, cloud resources, and recovery controls.

Source

Microsoft still cannot give administrators a release date for Exchange Server Subscription Edition (SE) Cumulative Update 1 (CU1), saying an AI-assisted security push has consumed the Exchange team’s testing capacity.

Source

Microsoft is urging IT administrators to make Microsoft Entra ID the strategic identity platform for modern workloads instead of relying exclusively on on-premises Active Directory. The recommended path is incremental: move authentication, access decisions, and governance to the cloud while retaining AD only where legacy applications and devices still require it.

Source

Microsoft’s August 2026 update extends Windows 11’s Low Latency Profile CPU boost beyond the shell to third-party Win32 apps, making launches noticeably faster on systems that receive the controlled rollout. Microsoft’s earlier K2 push targeted shell responsiveness and memory efficiency, as covered in earlier coverage. The change is delivered through KB5121003 and can affect how users experience application startup across managed PCs.

Source

Microsoft Purview is the Microsoft 365 tool for finding sensitive information in work emails and files, and then warning or blocking you if you try to share it outside the company or with someone who should not receive it. Data Loss Prevention (DLP) is a feature inside Microsoft Purview that checks work emails, files, chats, and some AI prompts for sensitive information and can stop it from being shared improperly. Microsoft Purview is preparing to apply DLP and auto-labeling to Google Workspace, Box, Dropbox, Salesforce, and other non-Microsoft platforms, but the rollout requires connectors, enterprise licensing, and additional storage charges. Earlier Purview coverage focused on restricting external sharing in Microsoft 365.

Source

Microsoft confirms a bug in Classic Outlook for Windows that removes Copilot Chat and its entry points after build 20026.20182 or later, although some users may welcome the missing AI controls. Earlier Outlook coverage has tracked Microsoft’s broader feature and Copilot changes, including the planned automatic Copilot rollout.

Source

Matthew Berman says Grok 4.6 is xAI’s biggest step forward yet, combining strong coding and knowledge-work performance with lower pricing than leading competitors. The model is available through Cursor, Grok Build, its API, OpenRouter, Vercel, and Cloudflare, while Grok 4.7 is reportedly only weeks away.

Source

Azure Private Link over IPv6 is a public preview that lets an IPv6 client reach selected Azure platform-as-a-service (PaaS) resources through a private endpoint. A private endpoint is a network interface with a private address in your Azure virtual network that represents a specific service resource. This article explains what PaaS means, how private access worked before this preview, the practical benefits of IPv6, and the configuration requirements and preview limitations you need to consider.

Source

OpenAI is previewing GPT-5.6 Sol Ultrafast, an API service tier that uses Cerebras hardware to deliver up to 750 output tokens per second—14 times faster than Standard processing. The limited rollout targets latency-sensitive work such as incident response, where teams need to analyze logs, traces, code changes, and reports while an outage is still developing. OpenAI has not disclosed separate pricing for Ultrafast; the standard GPT-5.6 Sol API rate is listed at $5 per million input tokens and $30 per million output tokens.

Source

Our earlier analysis of DeepSeek V4-Flash 0731 examined the company’s low-cost model alongside OpenAI’s GPT-5.6 Luna. DeepSeek has now released the new V4-Pro-0813, an upgraded flagship that delivers slightly lower performance than OpenAI’s GPT-5.6 Sol on broad intelligence benchmarks while costing substantially less. The release also introduces the MIT-licensed DeepSeek Harness agent framework.

Source