X significantly expanded its open-source initiative by releasing the complete source code for its "For You" ranking engine and core filtering systems on GitHub under the Apache v2 license.

Simultaneously, the platform launched a new "Under the Hood" transparency tool within app settings, allowing users to download a JSON file revealing if their account or posts have been assigned visibility-limiting labels (often referred to as shadowbans) over the past month.

Thursday's release includes model configurations, filter logic, and ranking parameters, enabling developers to run the Phoenix scoring system independently to verify how content is suppressed or promoted.

The open-sourced code confirms that the "For You" feed relies on a Grok-based transformer model to rank content, blending roughly 50% in-network posts (from followed accounts) with 50% out-of-network recommendations.

The 2026 update reveals a strict engagement hierarchy where Replies carry the most weight, followed by Reposts, then Likes, while posts containing external links are systematically deprioritized to keep users on the platform. Additionally, the code exposes a visibility multiplier for X Premium subscribers, which boosts their distribution frequency and position in feeds compared to non-paying users.

The new shadowban detection feature is initially piloting with accounts older than one year that have posted at least 10 times in the last month, providing raw data that users can analyze via LLMs against the public codebase.

Elon Musk and X's VP of Product Keith Coleman framed this move as a method to let users verify fairness themselves, with Coleman noting that external researchers have already successfully trained and run the scoring system using the provided code. The repository allows for community pull requests to suggest improvements.

"In making 𝕏 open source, we are actively seeking criticism to improve fairness and quality of algorithmic recommendations," Musk wrote on X.

The CIA and other U.S. intelligence officials held "low confidence" in Israeli-derived intelligence warning of an Iranian assassination plot against President Donald Trump, yet the Secret Service executed a covert plane swap in Turkey on July 8, 2026, as a precaution.

The incident occurred following the NATO Summit in Ankara, where Trump was secretly moved from a legacy Air Force One to a smaller C-32A military aircraft using a catering truck for transfer, while the original plane departed as a decoy.

CIA analysts described the threat as "Israeli-derived, not U.S.-generated" and deemed the information "not compelling enough" to confirm a specific danger. Some U.S. officials suspect the warning may have been intended to influence Trump's decision-making regarding Middle East policy rather than solely to inform him of a genuine threat.

Despite this skepticism, the Secret Service proceeded with the elaborate deception operation due to the high stakes involved, citing three previous near-miss assassination attempts against the president during his campaign and term. Secretary of State Marco Rubio was briefed on both the intelligence and the operational plan but reportedly remained on the decoy aircraft.

This episode highlights a growing divergence in intelligence assessment between the U.S. and Israel, with former officials noting a pattern where Israeli reporting is sometimes viewed as designed to shape presidential actions. The White House has declined to detail the specific intelligence, saying only that the president faces numerous threats.

SpaceXAI (formerly xAI and Cursor) launched Grok Bot in early beta late Tuesday, as a joint product preceding the full merger of the two companies. With an impressive slick UX design, Grok Bot acts as an enterprise-focused suite of persistent, always-on AI agents designed to act as digital coworkers rather than traditional chat prompts.

Users interact with Grok Bots as they would a human coworker via desktop or mobile apps.

Every Grok Bot gets its own cloud-hosted computer environment (a virtual machine). This allows them to operate 24/7 without stalling or relying on your local machine being powered on. Equipped with a browser, filesystem, and terminal, instead of relying strictly on formal APIs or webhooks, the bots can sign into third-party apps, emails, and websites just like a human user.

Bots complete multi-step workflows end-to-end without requiring clean APIs or MCP integrations, returning only when human approval is needed. Users can complete a task once while letting a Grok Bot observe the workflow. The bot saves the procedure as a repeatable routine, learn user preferences (such as writing style), and executes it autonomously on a schedule or trigger.

Users can deploy multiple specialized bots that share the same cloud environment, messaging each other in threads or group chats to handle complex, parallel tasks such as a Chief of Staff bot managing specialized lane bots for sales outreach, bug triage, expenses, or operations. The bots pass context and coordinate among themselves, bringing humans in only for final sign-offs or high-level judgment calls.

Unlike traditional automation tools that require complex setup, Grok Bot allows users to assign tasks via natural language chat, leveraging its ability to learn from live demonstrations of user behavior.

Initial access is restricted to high-tier subscribers of SuperGrok Heavy, Cursor Ultra, and Cursor Teams Premium. The system is available on macOS, Windows, Linux, and iOS (Android support upcoming), with an enterprise waitlist available for broader organizational deployment. 

Grok Bot was originally developed internally at SpaceXAI to automate routine internal operations, including engineering bug repros, marketing campaigns, customer support, and sales pipeline management. The release represents the company's strategic push beyond consumer LLM chat (Grok models) into enterprise autonomous agent execution—often referred to as "digital labor."