Moonshot AI has unveiled Kimi K3, a massive 2.8 trillion-parameter model that currently stands as the largest open-weight artificial intelligence system available. The model utilizes a mixture-of-experts architecture with 896 experts, though it only activates 16 experts per token to maintain computational efficiency. It features a 1 million-token context window and native vision support, allowing it to process vast amounts of data and visual information in a single prompt.

Source

GitHub has introduced a new moderation feature that allows repository administrators to archive pull requests instead of permanently deleting them. This capability is designed to help maintainers manage low-quality, spammy, or abusive contributions while preserving the data for internal records. When an administrator archives a pull request, the system automatically closes and locks the item to prevent further interaction.

Source

Moonshot AI has announced Kimi K3, a massive large language model featuring nearly three trillion parameters. This release marks a significant milestone as the first open-weight system to reach this scale, challenging the dominance of proprietary models. The model is specifically designed to handle high-end tasks including complex software development, multi-step reasoning, and advanced research.

Source

Microsoft has initiated the development cycle for Visual Studio Code version 1.130 with the release of the first Insiders build. This early version serves as a testing ground for upcoming features and stability improvements before they reach the stable branch. The initial release notes currently function as a placeholder while the development team begins merging new code contributions.

Source

Visual Studio Code version 1.99, also referred to as version 1.129 in some release cycles, introduces a significant "Agent Mode" for AI-driven development. This feature allows users to utilize autonomous agents that can fetch web content, find symbol references, and perform deep reasoning tasks. The update also integrates the Model Context Protocol (MCP), enabling these agents to connect with various external server tools for expanded functionality.

Source

OpenAI has launched its first physical hardware product, the Codex Micro, a specialized macropad designed to control AI-driven coding workflows. Developed in partnership with boutique manufacturer Work Louder, the $230 device is a modified version of the existing Creator Micro 2 platform. It features 13 low-profile mechanical keys, a rotary dial, and a joystick housed within a CNC-machined aluminum and polycarbonate body.

Source

A new class of cyberattack called Agent Data Injection (ADI) allows adversaries to manipulate AI agents by corrupting the factual context they rely on rather than hijacking direct instructions. Unlike traditional prompt injection, which embeds malicious commands, ADI exploits the way Large Language Models (LLMs) interpret data structures like sender names, button IDs, and tool logs. By inserting "probabilistic delimiters" such as fake quotes or brackets, attackers can trick an agent into perceiving fabricated metadata as legitimate system-level information.

Source

Visual Studio Code version 1.129 introduces a dedicated agent host process designed to run multiple AI models like GitHub Copilot, Anthropic Claude, and OpenAI Codex simultaneously. This background process operates independently of the main editor window, allowing AI sessions to persist even if the user closes the application. By moving these agents to a separate harness, developers can connect a single AI session across multiple editor windows to maintain a consistent workflow.

Source

xAI has released the full source code for its Grok Build terminal coding agent under the Apache 2.0 license following a significant privacy breach. Security researchers discovered that the tool was silently exfiltrating entire local Git repositories, including sensitive SSH keys, credentials, and full commit histories, to xAI’s Google Cloud storage. The incident prompted a public outcry, leading Elon Musk to pledge the deletion of all uploaded user data and the transition of the tool to an open-source model.

Source

Atlassian has updated Jira to better support AI-native software development by integrating its Teamwork Graph data layer. This system connects organizational data including team structures, project goals, and existing code repositories to provide AI agents with necessary enterprise context. By linking these disparate data points, the platform aims to resolve productivity bottlenecks where AI tools previously lacked the background information to produce accurate results.

Source

OpenAI has updated its Codex AI agent to encrypt messages sent between primary agents and sub-agents using the MultiAgentV2 protocol. This security measure ensures that sensitive task instructions are encrypted on the backend and never reach the local machine in clear-text format. While this prevents unauthorized interception or local tampering, it simultaneously removes the human-readable audit trail previously available to developers and administrators.

Source

GitHub Copilot in Visual Studio 2026 now includes a dedicated usage window to help developers manage usage-based billing models. This interface provides real-time consumption updates and allows users to set custom notification thresholds for approaching credit limits. These proactive alerts ensure that teams can monitor overages and maintain budget control directly within the development environment.

Source

GitHub Copilot for JetBrains now supports a "bring-your-own-key" (BYOK) model that allows users to connect to OpenAI-compatible custom endpoints. This update enables developers to use external Large Language Model (LLM) providers such as Anthropic, Azure OpenAI, or locally hosted models via Ollama without a paid Copilot subscription. By decoupling the agent runtime from the model subscription, GitHub allows teams to utilize its task orchestration infrastructure while paying model providers directly.

Source

GitHub has introduced AI-powered security detections that appear directly within pull request reviews to identify vulnerabilities before code is merged. This feature expands scanning capabilities to include programming languages and frameworks that are not currently supported by the standard CodeQL analysis engine. These AI-generated findings are labeled specifically to distinguish them from traditional static analysis results, ensuring developers can identify the source of the alert.

Source

Anthropic has confirmed that malicious actors, including state-sponsored groups, have successfully weaponized the Claude Code AI assistant to conduct highly automated cyberattacks. Vulnerabilities in versions 2.1.91 through 2.1.196 allow attackers to execute remote code and harvest sensitive data by injecting malicious prompts into source files like README.md. These exploits leverage the tool's "auto-mode" to bypass traditional security rules, performing unauthorized actions such as querying databases or calling APIs without human intervention.

Source

Oracle has updated its Fusion AI Agent Studio to support professional developers by integrating pro-code tools into the existing low-code framework. This expansion allows developers to build agentic applications using familiar environments such as Visual Studio Code, command-line interfaces, and Git-based workflows. The platform now supports AI coding assistants like Anthropic Claude Code and OpenAI Codex to help generate and validate Fusion-specific artifacts.

Source

Anthropic has introduced a new tokenizer for its Claude Sonnet 5 and Opus models that significantly increases the number of tokens generated for the same input text. Tokenizers are essential components that translate human language or code into numerical integers that large language models can process. Because AI providers bill based on token volume, this architectural change directly impacts the total cost of ownership for organizations utilizing these models.

Source

The IETF has officially standardized the HTTP QUERY method under RFC 10008 to address long-standing limitations in web communication. This new verb serves as a middle ground between GET and POST by allowing request data to be sent in the body rather than the URL. It is specifically designed for complex search operations that require structured data but do not change the state of the server.

Source

Cynative is an open-source security research agent designed to identify vulnerabilities across cloud environments, Kubernetes clusters, and code repositories. The tool functions by orchestrating large language models to analyze infrastructure and runtime data in an adversarial manner similar to a human attacker. It supports major platforms including AWS, Azure, and Google Cloud, utilizing existing credentials found in the operator's environment to perform its investigations.

Source