AI News — Saturday, March 28, 2026

A new $40 billion loan to SoftBank is sparking rumors of a potential OpenAI IPO in 2026, indicating significant financial movements in the AI industry.

Researchers introduce Intern-S1-Pro, a groundbreaking scientific multimodal foundation model operating at a trillion-parameter scale, promising significant advancements in scientific AI.

OpenAI announces the acquisition of Astral, signaling a strategic move to enhance its technological stack and market position in the rapidly evolving AI landscape.

Venture capitalists are pouring billions into the next generation of AI, raising questions about OpenAI's strategic direction following reports of Sora's potential discontinuation.

New research introduces PixelSmile, a novel approach aimed at achieving more precise and fine-grained control over facial expression editing in images.

Google is rolling out its 'Personal Intelligence' features to a wider global audience, aiming to make advanced AI capabilities more accessible to everyday users.

Listen Labs has successfully raised $69 million in funding to expand its AI-powered customer interview platform, following a highly publicized viral billboard hiring campaign.

ARC-AGI-3 is introduced as a new and more complex challenge designed to push the boundaries and evaluate the capabilities of frontier agentic artificial intelligence systems.

OpenAI shares insights into its methods for monitoring internal coding agents, focusing on strategies to detect and prevent potential misalignment in their behavior and outputs.

A new paper introduces RealRestorer, a method that leverages large-scale image editing models to achieve more generalizable and effective restoration of real-world images.

Calibri presents a new technique for parameter-efficient calibration that significantly improves the performance and efficiency of diffusion transformers.

Cursor is under fire for allegedly using the Chinese AI model Kimi K2.5 without proper disclosure, raising concerns about transparency and trust among developers.

UI-Voyager introduces a novel self-evolving GUI agent that improves its capabilities by learning directly from its failed interactions and experiences.

A technical guide demonstrates how to construct more reliable AI agents by implementing the transactional outbox pattern alongside Redis Streams for robust message handling.

A developer shares how they built a three-agent AI pipeline that automatically converts weekly GitHub activity into structured blog posts for Notion and DEV.to.