AI News — Friday, May 29, 2026
Google announced a hundred new features and updates across its products and platforms at I/O 2026, with a strong focus on advancements in AI.
OpenAI has published its new Frontier Governance Framework, outlining principles and practices for the safe and responsible development of highly advanced AI models.
Work management platform Asana has acquired StackAI, a no-code platform for building AI agents, signaling a move to integrate more advanced AI capabilities into its offerings.
Railway has raised $100 million in funding to develop an AI-native cloud infrastructure, positioning itself as a direct competitor to established cloud providers like AWS.
Google has unveiled the first redesign of its search box in 25 years, a change that is expected to signify deeper integration of AI into its core search experience.
Glean, an enterprise AI search and knowledge discovery platform, has surpassed $300 million in revenue, driven by its ability to help companies cut AI-related costs.
This article explores how the underlying architecture and functionality of the internet are evolving to better serve the needs of AI agents and automated systems.
Researchers introduce ProRL, a new reinforcement learning method that uses rectified policy gradient estimation to significantly improve proactive recommendation systems.
A new research paper details Agent Explorative Policy Optimization, a technique designed to enhance multimodal agentic reasoning by enabling more effective exploration strategies.
OpenAI has released its strategy and safeguards for managing election-related information and preventing misuse of its AI models during the 2026 election cycle.
DenoiseRL presents a novel method for bootstrapping reasoning models, allowing them to effectively recover and perform accurately even when faced with noisy or incomplete input prefixes.
New research demonstrates that Generative Supervision (GEM) significantly aids the development and capabilities of embodied AI, pushing the boundaries of intelligent agents interacting with the physical world.
MemTrace introduces a new system for tracing and attributing errors within the complex memory systems of large language models, crucial for debugging and improving LLM reliability.
This paper proposes a method for automated domain specialization, enabling small computer-use agents to learn from their weaknesses and become more efficient in specific tasks.
A developer shares their experience highlighting the significant challenges and time investment involved in debugging AI-driven code compared to its initial development.