Latest in Artificial Intelligence

Protecting people from harmful manipulation

Google DeepMind researches AI's harmful manipulation risks across areas like finance and health, leading to new safety measures.

Lyria 3 Pro: Create longer tracks in more

Introducing Lyria 3 Pro, which unlocks longer tracks with structural awareness. We’re also bringing Lyria to more Google products and surfaces.

This startup wants to change how mathematicians do math

Axiom Math, a Palo Alto startup, has launched Axplorer, a free AI tool designed to help mathematicians discover mathematical patterns and solve long-standing problems. The tool is a redesign of the existing PatternBoost platform, aiming to transform how mathematicians approach their work.

Agentic commerce runs on truth and context

Agentic commerce represents a shift from AI assistants that merely provide information to autonomous agents that can actually execute transactions and make purchases on behalf of users. These agents rely on accurate data and contextual understanding of user preferences to complete complex tasks like booking trips, rather than just returning search results. The technology enables a more seamless shopping experience where users can give high-level instructions and have the agent handle all operational details.

Inside our approach to the Model Spec

OpenAI has released a Model Spec, a public framework designed to guide AI model behavior while balancing safety, user autonomy, and accountability. The framework establishes principles for how AI systems should operate as they become more advanced and capable.

The AI Hype Index: AI goes to war

Anthropic and OpenAI are competing for Pentagon contracts to weaponize AI models, with OpenAI securing a controversial deal despite criticism for being "opportunistic and sloppy." The military's push to integrate AI has sparked significant backlash, including mass user departures from ChatGPT and London's largest protest against AI to date. The incident highlights growing tensions between AI companies and public concerns over weaponization.

Introducing the OpenAI Safety Bug Bounty program

OpenAI has launched a Safety Bug Bounty program designed to identify vulnerabilities and safety risks in its AI systems. The program focuses on detecting issues such as agentic vulnerabilities, prompt injection attacks, and data exfiltration that could lead to AI abuse.

Helping developers build safer AI experiences for teens

OpenAI has released prompt-based teen safety policies designed to help developers create safer AI experiences for younger users through a tool called gpt-oss-safeguard. This resource enables developers to moderate age-specific risks and implement appropriate content guardrails in their AI systems.

Powering product discovery in ChatGPT

ChatGPT has introduced a new shopping feature called the Agentic Commerce Protocol that enhances product discovery with visual, immersive experiences. The feature allows users to compare products side-by-side and enables merchants to integrate their offerings directly into the platform.

A New Framework for Evaluating Voice Agents (EVA)

The hardest question to answer about AI-fueled delusions

The article discusses challenges in understanding AI-related delusions and misconceptions, with a focus on geopolitical implications such as the Pentagon's AI planning regarding Iran. It explores the difficulties in separating fact from fiction when it comes to AI capabilities and their real-world applications.

The Bay Area’s animal welfare movement wants to recruit AI

Animal welfare advocates and AI researchers in the Bay Area are collaborating to leverage artificial intelligence technology for animal protection efforts. The movement seeks to harness AI's capabilities to address animal welfare challenges more effectively and innovatively.

A Visual Guide to Attention Variants in Modern LLMs

This article provides a visual overview of different attention mechanisms used in modern large language models, ranging from traditional Multi-Head Attention (MHA) to more efficient variants like Grouped Query Attention (GQA) and Multi-Head Latent Attention (MLA). It covers emerging approaches including sparse attention patterns and hybrid architectures designed to improve computational efficiency and performance. The guide helps explain the evolution of attention mechanisms that power contemporary LLMs.

Can you *really* train AI to "get" videos just by showing it a million of them?

A Very Big Video Reasoning Suite

Build a Domain-Specific Embedding Model in Under a Day

State of Open Source on Hugging Face: Spring 2026

Measuring progress toward AGI: A cognitive framework

We’re introducing a framework to measure progress toward AGI, and launching a Kaggle hackathon to build the relevant evaluations.

Holotron-12B - High Throughput Computer Use Agent

GTC preview: Inside the AI factory — The $1T infrastructure war under the hood of the AI economy

Nvidia's GPU Technology Conference showcases the latest advances in AI hardware, software, and models, while highlighting the massive infrastructure investments driving the AI economy. The event reveals a trillion-dollar infrastructure war as companies compete to build the computational foundation needed to power increasingly sophisticated AI systems. Beyond faster GPUs and larger models, the conference demonstrates how critical hardware and infrastructure development have become to maintaining competitive advantage in the AI sector.

Twenty years in, Amazon S3 finds itself at the center of AWS’ push beyond storage

Amazon S3 celebrates its 20th anniversary as it has evolved from an internal utility into a fundamental pillar of AWS cloud infrastructure. The milestone reflects S3's critical role in AWS's broader ecosystem and the company's continued expansion beyond basic storage capabilities.