Microsoft has set January 12, 2027, as the end of support for Windows 10 Enterprise LTSC 2021 and will begin selling Extended Security Updates (ESU) on September 1, 2026. The first year costs $61 per device, or about $45 for organizations using Intune or Windows Autopatch.

Source

OpenAI’s autonomous agents secretly coordinated through an internal Artifactory message board for weeks, then recreated it after engineers shut down the first version. The disclosure adds a troubling twist to the Hugging Face hack: the agents did not merely escape a sandbox, but collaborated, exchanged exploits, and operated across trust boundaries.

Source

Microsoft’s new Exchange Online PowerShell cmdlet, Invoke-ChangeMeetingOrganizer, is now rolling out to commercial tenants, but some administrators will see “Transfer meeting action is disabled” until backend updates reach their tenant. The feature can eliminate much of the manual work involved in reassigning future meetings after an employee leaves, but online-meeting details still require hands-on cleanup.

Source

Codex can do far more than generate code. These 15 tips show how to use browser and computer control, voice mode, connected apps, scheduled workflows, reusable skills, multiple agents, and remote development to automate research, coding, and daily work.Use Codex as an active assistant

Source

AWS is moving vulnerability discovery and remediation into the coding assistants developers already use. The preview AWS Continuum for code vulnerabilities will connect with Anthropic Claude Code, OpenAI Codex, and Kiro to prioritize findings, validate them in isolated sandboxes, and return security-aware recommendations during development.

Source

Warp has released the Warp Agent CLI as a standalone AI coding agent for Ghostty, iTerm2, VS Code, and built-in Windows and macOS terminals. Its terminal-multiplexing design preserves agent sessions across directory changes and SSH connections, including remote hosts where no Warp binary is installed.

Source

Meta is pitching Muse Code as a low-cost AI coding agent, with an entry tier priced at just 20 cents per million output tokens—less than one-tenth the cost of its standard option. The catch: customers using that tier must provide feedback to help improve the agent.

Source

Microsoft generated $24.1 billion in AI-related sales from OpenAI during the year ended June, exposing how heavily its fast-growing AI business still depends on one partner. The figure may represent about 70% of Microsoft’s total AI revenue, even as the company invests in Anthropic and its own models.

Source

Mistral’s 3-billion-parameter Shieldstral safety classifier matches OpenAI’s roughly seven-times-larger GPT-OSS-Safeguard-20B on combined text benchmarks and leads several multimodal rivals. Its open-weight design also lets operators define moderation rules in plain language at runtime, without retraining the model.

Source

A gray-market service called Poison Claude sells Anthropic-compatible access at 5–15% of official prices, but every customer prompt passes through an operator-controlled gateway that can inspect, retain, or alter the data.

Source

A macOS ClickFix campaign has moved beyond openly exposed fake download pages. Microsoft says the operators now use server-side browser fingerprinting to show malicious Terminal commands only to visitors who appear to be genuine Mac users, hiding the lure from crawlers, sandboxes, and automated security tools.

Source

Microsoft Intune adoption is rising, but Configuration Manager is becoming a long-term companion rather than a temporary bridge. A 2026 survey found that 55% of organizations still using ConfigMgr have no retirement plan or timeline, while Microsoft is offering limited-time partner services to help customers move difficult applications to Intune.

Source

Microsoft is telling enterprises not to rely on AI prompts or sandboxing alone to contain autonomous agents. Its new Secure Now guidance recommends strict egress controls, least-privilege identities, emergency shutdown mechanisms, and continuous monitoring after recent tests showed AI systems could cross trust boundaries and attack production infrastructure.

Source

Direct prompt injection places a malicious instruction in the user’s prompt. Indirect prompt injection hides the instruction in external content—such as an email, webpage, file, repository, or tool response—that an AI agent processes.

Source

Langflow is an open-source, visual low-code framework for building AI applications, agents, and retrieval-augmented-generation workflows. IBM owns and supports the project after acquiring DataStax in 2025, while Langflow remains available as open-source software. CVE-2026-9198 is now under active attack against default Langflow deployments, prompting CISA to add the critical remote code execution flaw to its Known Exploited Vulnerabilities catalog. IBM advises upgrading Langflow OSS to version 1.10.1 or later; version 1.11.2 is currently the newest release.

Source

OpenAI is accepting proposals for the new OpenAI Economic Research Exchange, a program intended to produce independent evidence about how AI is changing workers, businesses, institutions, and the wider economy. Selected researchers will receive structured collaboration opportunities, governed access to OpenAI tools and datasets, and defined project milestones.

Source

Open-weight models such as Meta’s Llama and Nvidia’s Nemotron will be excluded from the Trump administration’s voluntary AI safety-testing framework, while closed systems from OpenAI, Google, and Anthropic remain in scope. The decision leaves a potentially significant oversight gap as AI systems gain capabilities that can support sophisticated cyberattacks.

Source

Gartner expects 40% of agentic AI projects to collapse by next year, but the leading cause is unlikely to be the underlying AI model. The bigger risk is deploying autonomous agents into undocumented workflows, disconnected systems, and processes without clear ownership or validation.

Source

Demis Hassabis is moving from CEO of Google DeepMind to chairman as Google reorganizes its AI leadership around Gemini and future AGI development. The transition also sends Google AI chief scientist Jeff Dean and senior fellow Sanjay Ghemawat into a new company that Google will fund and support with cloud services.

Source