AI News — Friday, April 3, 2026
Reports indicate a significant leak of Anthropic Claude's source code, raising major security and competitive concerns within the AI industry.
Microsoft has unveiled three new foundational AI models, intensifying its competition with other major players in the rapidly evolving AI landscape.
Anthropic introduces 'Cowork,' a new desktop agent for Claude designed to integrate directly with user files and automate tasks without requiring coding skills.
OpenAI has acquired TBPN, a popular founder-led business talk show, signaling a potential expansion into media and content creation for the AI giant.
Nous Research releases NousCoder-14B, an open-source coding model, strategically timed with recent events surrounding Claude's code, offering an alternative for developers.
Railway has raised $100 million to develop AI-native cloud infrastructure, positioning itself as a significant challenger to established cloud providers like AWS.
Google's Vids app now allows users to control avatars using natural language prompts, enhancing its AI-powered video creation capabilities.
A new paper introduces ClawKeeper, a robust framework designed to provide comprehensive safety and security for OpenClaw AI agents using a multi-layered approach.
OpenAI has launched a new bug bounty program focused on safety, inviting researchers to identify and report vulnerabilities in its AI systems.
New research suggests that terminal-based AI agents are highly effective and sufficient for automating a wide range of tasks within enterprise environments.
Google introduces new options for the Gemini API, allowing developers to better balance inference cost and reliability based on their specific application needs.
A mechanistic study explores how the simulation or understanding of emotions can influence the behavior and decision-making processes of large language models and AI agents.
Researchers propose a community-driven framework aimed at developing open, reliable, and collaborative tool-using AI agents, fostering broader participation and trust.
A new paper investigates how contextual information can subtly alter and shorten the reasoning paths taken by large language models, impacting their output.
Introducing Dynin-Omni, a novel omnimodal unified large diffusion language model capable of processing and generating across various data modalities.