AI Daily: X Platform Promises Full Open Source, OpenAI Releases Physical Keyboard 'Codex Micro', SpaceXAI Releases Grok Build Source Code

AI Daily: X Platform Promises Full Open Source, OpenAI Releases Physical Keyboard ‘Codex Micro’, SpaceXAI Releases Grok Build Source Code

Today’s headline news covers a total transparency plan for a social platform, physical hardware innovation, and comprehensive upgrades for major AI tools, demonstrating that cutting-edge technology has thoroughly integrated into daily development and life. Here is a curated list of today’s hot topics and new tool applications for you.

1. Elon Musk Announces X Platform Code Will Be Fully Open-Source

Elon Musk recently promised in an X platform tweet that once the security vulnerability review is complete, X platform’s full source code will be released without reservation. To ensure total transparency, the official team will invite third-party auditors to verify the running system, comparing the open-source code with the actual system in operation. In the future, anyone will be able to verify the algorithms running on the X platform themselves. Building trust through complete transparency is an effective solution to eliminate concerns about algorithmic bias, holding high symbolic significance for the entire tech industry.

Once we have completed our review for security vulnerabilities, we will make the entire codebase of 𝕏 open source, with no exceptions.

Moreover, we will invite third party reviewers to examine the system that is running to confirm that the open source code is what is running.…

— Elon Musk (@elonmusk) July 15, 2026

2. xAI Releases Grok Build, Crafting a Local-First Development Experience

The xAI team officially announced Grok Build is open-source. This code agent and terminal user interface belonging to SpaceXAI is now available on GitHub.

The biggest technical highlight of Grok Build is its “local-first” execution capability. Developers only need to compile it themselves and point it to a local inference model, driving the entire system via a simple config.toml file. The open-source content includes agent loops, various practical tools, and a terminal interface responsible for rendering and input handling. For developers currently researching expansion systems (such as skills, plugins, MCP servers, or sub-agents), this source code is an excellent reference.

3. OpenAI Crosses Over to Release Codex Micro, a $230 Physical Control Center

OpenAI and Work Louder have partnered to launch a small keyboard named Codex Micro in the Supply Co. store. The device, priced at $230, is positioned by the official team as a “command center for agentic work.”

In terms of hardware design, the keyboard is equipped with 13 mechanical switches, paired with 1 touch sensor, 1 rotary encoder, and 1 planar joystick. The keys use RGB lighting to reflect the assistant’s real-time state (such as idle, thinking, or executing). Users can quickly launch code review or debugging using the joystick, and utilize the knob to dynamically adjust the model’s reasoning intensity (thinking strength).

This product has sparked polarized discussions on the Reddit forum. Some netizens believe the $230 price is too high, poking fun that the functionality could be perfectly replaced by AutoHotKey. Furthermore, a netizen sarcastically commented that the product looks like “they hired Jony Ive (Apple’s former Chief Design Officer) to design it,” to which other netizens hilariously replied: “Is that the same guy who designed the most beautiful Ferrari in history?”

Nevertheless, some developers still expressed support for the physical tactile feedback, even planning to use it in conjunction with Apple Vision Pro; the community has already begun researching workflows to bind Claude and other tools to this device.

4. GPT-Red Debuts, Automated Red-Teaming Improves Model Defense

OpenAI recently released detailed information about GPT-Red, a powerful model focused on automated red-teaming. Traditional red-teaming relying on humans is time-consuming and hard to scale; GPT-Red is trained through self-play reinforcement learning mechanisms, continuously evolving during offensive and defensive processes, effectively identifying system vulnerabilities and patching them before deployment.

Actually applied to the training of the latest model GPT-5.6 Sol, GPT-Red successfully caused the failure rate of prompt injection attacks to drop significantly. For example, the success rate of “jailbreak” attacks on GPT-5.1 was as high as 95%, while it has dropped to below 10% on GPT-5.6 Sol, and the failure rate against direct prompt injection is only 0.05%. The official team also tested GPT-Red on a physical vending machine system called Vendy, successfully tampering with high-price item prices to $0.5 and canceling others’ orders, confirming the urgency of introducing automated defense mechanisms into production environments.

5. Claude Code Artifacts Now Supports MCP Connector

According to the latest update from ClaudeDevs, Claude Code artifacts now officially support calling MCP connectors. Through this, developers can create dashboards and applications that grab external information in real-time; the system can dynamically obtain data and execute actions based on user requirements.

This feature is currently available to users on Pro, Max, Team, and Enterprise plans, but does not apply to artifacts shared publicly. This expansion significantly enhances Claude’s flexibility in enterprise and commercial applications.

Claude Code artifacts can now call MCP connectors, letting you build dashboards and apps that can fetch information and take actions for each viewer on demand.

Available on Pro, Max, Team, and Enterprise plans. Not available on publicly-shared artifacts. pic.twitter.com/FcvtiOqv0J

— ClaudeDevs (@ClaudeDevs) July 15, 2026

6. Cursor Adds More, Significantly Increasing Model Usage Quotas

Well-known developer Lee Robinson announced on Twitter that Cursor has fully doubled the included model usage across all plans. This update allows users to access powerful tools like Grok 4.5 and Composer 2.5 more frequently; in a fiercely competitive landscape of development tools, strategies that give direct feedback to users help further consolidate the developer ecosystem.

We just doubled the included usage of Cursor models on all plans.

Enjoy more access to Grok 4.5 and Composer 2.5!

— Lee Robinson (@leerob) July 16, 2026

7. GPT-5.6 Experience Sharing Activity Ends, Community Enthusiasm Remains High

The GPT-5.6 experience sharing activity previously held by OpenAI has officially closed its submission channel, and the official team is conducting subsequent reviews. This event attracted a large number of ChatGPT Go, Plus, and Pro subscribers to participate; as long as they publicly shared their compliant, authentic experiences on the X (Twitter) platform, the first 10,000 qualified participants will receive a ChatGPT point reward worth $100 (approximately 2,500 points).

💡 Author’s Perspective on Today’s Highlights:

The most interesting thing today is OpenAI launching the Codex Micro physical keyboard. Although it is generally considered on Reddit to be a “faith tax” at $230, it truly reveals a new hardware trend in the AI era—physicalization and tactile feedback. With the advent of the Agent era, we might see more such physical peripherals designed specifically for AI collaboration in the future.

Q&A

Q1: What specific promises did Elon Musk make regarding the X platform’s open-source plan? A: Elon Musk stated that after the security vulnerability review is completed, X platform’s full source code will be released “with no exceptions.” To establish complete transparency and trust, the official team will even invite third-party auditors to verify the system in operation to confirm that the open-sourced code is identical to the actual running system.

Q2: What features does Grok Build, open-sourced by SpaceXAI, have? A: Grok Build is a terminal interface (TUI) AI code agent tool. It focuses on full-screen, mouse interaction, and high extensibility. Its biggest feature is its “local-first” execution capability; developers can compile it themselves, point it to a local inference model, and drive the entire system entirely through a config.toml configuration file. The source code also includes operational methods for expansion systems like agent loops, tool implementation, and MCP servers.

Q3: What are the specific specifications and functions of the physical keyboard Codex Micro launched by OpenAI? A: Codex Micro is a physical device co-designed by OpenAI and Work Louder, priced at $230, and positioned as a “command center for agentic work.” In terms of hardware, it is equipped with 13 mechanical switches, 1 touch sensor, 1 rotary encoder (knob), and 1 planar joystick. The keys reflect the AI’s state (such as idle, thinking) through RGB lighting in real-time. Users can quickly launch code reviews or debugging by flicking the joystick, or adjust the model’s “thinking intensity” by rotating the knob.

Q4: What is the newly released GPT-Red by OpenAI? How does it improve model security? A: GPT-Red is a powerful automated security “red-teaming” model, trained through self-play reinforcement learning mechanisms. Traditional red-teaming relying on humans is hard to scale, while GPT-Red can automatically identify system vulnerabilities and patch them before model deployment. In actual application, it helped GPT-5.6 Sol significantly drop the failure rate against direct prompt injection attacks to only 0.05%. It even successfully tampered with high-price item prices to $0.5 and canceled others’ orders during tests on a vending machine system (Vendy), confirming its powerful potential for automated attack and defense.

Q5: What are the updates for Claude Code Artifacts and Cursor regarding developers today? A:

• Claude Code Artifacts: Now officially supports calling MCP connectors, allowing developers to create dashboards that grab external information in real-time and dynamically take action. This feature is open to users on Pro, Max, Team, and Enterprise plans, but not applicable to artifacts shared publicly.
• Cursor: Lee Robinson announced that Cursor models included in all plans will be “doubled,” allowing developers to more frequently use tools like Grok 4.5 and Composer 2.5.