AI News — Friday, June 12, 2026
Jeff Bezos's new venture, Prometheus, has secured an astounding $12 billion in funding to develop an "artificial general engineer" focused on real-world physical applications.
OpenAI announced its acquisition of Ona, signaling a strategic move to integrate new capabilities or talent into its AI ecosystem.
New research proposes redesigning Mixture-of-Experts (MoE) routers using Manifold Power Iteration to potentially enhance the efficiency and performance of large language models.
This paper explores a novel approach to achieving generalist autonomous research by refining hypothesis trees, pushing the boundaries of AI's scientific discovery capabilities.
This survey provides a comprehensive overview of agentic environment engineering for large language models, covering modeling, synthesis, evaluation, and application to guide future research.
A new benchmark, Claw-SWE-Bench, has been introduced to rigorously evaluate OpenClaw-style AI agent harnesses on complex coding tasks, aiming to standardize performance measurement.
New research proposes moving beyond simple scalar rewards in reinforcement learning by internalizing reasoning into score distributions, aiming for more sophisticated agent training.
OpenAI reiterates its commitment to supporting Europe's efforts in establishing a trustworthy AI ecosystem, highlighting collaboration on ethical guidelines and regulatory frameworks.
HazelJS has announced the stable 1.0.0 release of its AI-native TypeScript framework, offering developers a robust toolset for building AI applications.
Researchers demonstrate that grammar-constrained decoding methods can be exploited to jailbreak large language models, compelling them to generate malicious code.
DeNovoSWE introduces a method for scaling long-horizon environments, enabling AI agents to generate entire software repositories from scratch, pushing the boundaries of autonomous software engineering.
A Google AI blog post reflects on a child's insightful questions about the evolution of coding, highlighting the rapid advancements and future implications of AI in software development.
Google details five layers of security within its AI Development Kit (ADK) designed to protect AI agents from prompt injection attacks, enhancing the robustness of AI applications.
An astrophysicist shares insights into how OpenAI's Codex is being utilized to assist in complex simulations of black holes, demonstrating AI's utility in advanced scientific research.
Chromiumfish offers a stealthy Chromium build integrated with a Playwright harness, providing developers with a powerful tool for automated browser testing and scraping in Python and Node.js.