Musk Manually Clears Grok Build User Data, Agentic Coding Trust Crisis Resolved

Recently, the AI coding tool Grok Build has been embroiled in a privacy scandal. According to foreign media reports, the tool was exposed for uploading users' code repository data without explicit authorization, causing great concern among developers worldwide. Faced with the escalating controversy, Elon Musk personally responded, promising to completely delete all historical user data.

The incident was triggered by a "phishing" test conducted by a security researcher. The researcher created a fake test repository, embedded unique markers, and monitored the data flow throughout. The results were shocking: even though users had turned off the "help improve the model" option, Grok Build silently packaged the entire repository and its modification history, transmitting it to a third-party bucket. The amount of data transmitted reached several GBs, including sensitive directories that may contain keys and configuration files.

This incident quickly caused panic within the developer community. Code repositories are the core assets of developers, and once leaked, they not only mean the loss of commercial secrets but also directly threaten the security of production environments. Many users changed all access keys overnight and even chose to uninstall the related tools to protect themselves.

In response to the accusations, the xAI team quickly took action, stopping the illegal upload behavior and officially launching a new feature aimed at protecting user privacy, allowing users to turn off data retention with one click and trace and delete uploaded data. In his response, Musk made a firm statement, saying "Zero anything whatsoever will remain" (not a single byte will remain), clearly indicating that all previously collected data would be thoroughly erased.

From developer doubts to official deletion, the entire crisis was resolved within 48 hours. Although the data has been promised to be deleted, this incident undoubtedly served as a warning to the Agentic Coding (agent-based coding) sector. As an AI assistant with the highest level of computer permissions, how to maintain the principle of "privacy first" while providing productivity has become an issue that all developer tool vendors must face head-on.

PixVerse Completes $439 Million C-Round Expansion Financing, Valuation Surges to $2 Billion

Competition in the AI video generation sector is intensifying. Singapore-based startup PixVerse has completed a $439 million C-round expansion financing, with its valuation exceeding $2 billion, signaling strong growth confidence. The company was co-founded in 2023 by Wang Changhu, former head of computer vision at ByteDance, and Jaden Xie, following a typical technology-driven development path.

New Species of Mobile Phones Emerge: StepXun Announces Native Intelligent Agent System and the First Phone STEPX Neo

StepXun held the "StepXun Terminal Brand and Next Generation Intelligent Agent Strategy Launch Event" on the evening of July 13, announcing three innovative achievements: the native intelligent agent system Step AOS, the large model-native AI terminal brand STEPX, and the first phone STEPX Neo. The core is Step AOS, designed to restructure hardware scheduling and system capability allocation for intelligent agents, building a new generation of intelligent agent interaction.

SenseNova-Vision by SenseTime Open-Sources a Unified Vision Large Model, Achieving Outstanding Performance in Four Core Visual Tasks with a Single Model

SenseTime has released and open-sourced the "SenseNova-Vision" vision large model, whose core lies in natively integrating vision capabilities into a general foundation model, breaking the fragmented mode of packaging expert models such as detection and segmentation. This model achieves dominant performance across multiple evaluation tasks in four fields, marking a critical upgrade in visual tasks toward a unified native architecture.

OpenAI Criticizes AI Evaluation Benchmark: 731 Questions, Nearly a Third Have Flaws. 8-Month Passing Rate Rises from 23% to 80%, Now Ineffective

OpenAI publicly questioned the SWE-Bench Pro benchmark, pointing out that about 30% of its 731 test tasks have evaluation flaws. The benchmark, launched by Scale AI, is an industry authority for measuring large model programming capabilities. However, OpenAI warned that the passing rate of cutting-edge models has surged from 23.3% to 80.3% within 8 months, which is unusually fast, indicating doubts about the reliability of the evaluation.

Musk Manually Clears Grok Build User Data, Agentic Coding Trust Crisis Resolved

Related Recommendations

Musk Admits That Grok Build Leaked User Code and Promises to Completely Erase All Historical Data Without Leaving a Single Byte

PixVerse Completes $439 Million C-Round Expansion Financing, Valuation Surges to $2 Billion

New Species of Mobile Phones Emerge: StepXun Announces Native Intelligent Agent System and the First Phone STEPX Neo

SenseNova-Vision by SenseTime Open-Sources a Unified Vision Large Model, Achieving Outstanding Performance in Four Core Visual Tasks with a Single Model

OpenAI Criticizes AI Evaluation Benchmark: 731 Questions, Nearly a Third Have Flaws. 8-Month Passing Rate Rises from 23% to 80%, Now Ineffective