SpaceXAI has open-sourced Grok Build, a terminal-based AI coding agent capable of codebase understanding, editing, and executing shell commands. The Rust source code has been made public.
A former Google DeepMind researcher has published an essay detailing their resignation following the company's unrestricted military AI contract with the U.S. Department of Defense, citing the collapse of 2018 pledges and ethical standards.
Cadence has introduced AuraStack, an agentic AI platform on Allegro AI Studio designed to streamline the design process for printed circuit boards and advanced packaging through automated orchestration and multiphysics analysis.
Alexander Mordvintsev of Google has published a conceptual paper on MorphoHDL, a hardware description language centered on recursive splitting and rewiring.
Researchers at XQL Systems have implemented automatic differentiation and neural networks in SQL, drawing on DataFusion's visitor pattern and JAX, and presented their work at the first DataFusion showcase.
A security researcher pointed out undisclosed repository uploads by the Grok Build CLI. xAI has since disabled the feature via a server-side flag without providing a public explanation.
An independent researcher analyzed xAI's Grok Build CLI and confirmed that codebases are transmitted and stored during normal use.
A comparative experiment was conducted where 12 models, including the three tiers of GPT-5.6 and Meta's Muse Spark 1.1, each generated four types of web applications five times.
Apple has filed a lawsuit in U.S. federal court against OpenAI and two former employees, alleging that trade secrets regarding unannounced technology were stolen.
Four models, including Grok 4.5, were given the same prompts to compare the quality of generated applications and inference costs. Grok 4.5 was fast and inexpensive, though it failed on the first attempt for a complex 3D animation.
OpenAI is reportedly in early discussions to transfer 5% of its shares to the US government, a move proposed by CEO Sam Altman as a means to publicly distribute the benefits of AI.
Google has released Nano Banana 2 Lite, the latest model in the Gemini Image series, which the company claims achieves high-speed image generation and editing at a significantly lower cost than existing models.
The Financial Times reported that Google has imposed limits on Meta's use of Gemini, causing delays in Meta's internal projects due to a lack of computing power.
NBC News reported that the U.S. government has permitted Anthropic to release Mythos AI to specific trusted organizations.
A preview of the next-generation model GPT-5.6 Sol has been presented, recording 1,139 points and 744 comments on Hacker News.
On June 24, 2026, OpenAI announced Jalapeno, its first custom chip co-developed with Broadcom, designed to improve the efficiency of AI inference.
Google DeepMind has announced the integration of Computer Use capabilities into Gemini 3.5 Flash, making it easier to build agents that operate across browsers, mobile, and desktop environments.
Reid Hoffman characterized SpaceX's AI strategy as securing relevance through acquisitions and evaluated xAI as a complete train wreck.
Sakana AI has launched Sakana Fugu, an API service that dynamically orchestrates top models from around the world to solve complex tasks. The company employs a mechanism that organizes agents through learning rather than relying on manual design.
Andrew Marble argues that the operational penalties for switching to open models remain minimal, citing factors such as the introduction of ID verification for Claude.