Frontier models, research results, agents, and developer tooling.
A new red-team study finds that leading text-to-image models from Google and OpenAI can be jailbroken with off-the-shelf tools, and newer versions are even more vulnerable.
Enterprises should shift from raw AI spending to measuring useful work per dollar as they scale agentic workflows.
A new Hugging Face tutorial demonstrates how to profile attention layers in PyTorch to identify performance bottlenecks.
A new generative-AI initiative seeks to turn raw wearable sensor streams into a general-purpose health intelligence layer.
OpenAI has published a public framework for government and national security partnerships, grounding the work in democratic accountability and public safety.
Google Research explores the algorithmic roots of creativity in diffusion models, shedding light on how they generate novel content.
OpenAI's case study shows how Deutsche Telekom is embedding AI across customer service, network operations, and employee workflows to become an AI-native telco.
A new voice-model generation signals OpenAI's push toward more natural, conversational AI interactions in ChatGPT.
OpenAI's latest model now powers Microsoft's office suite, promising faster and better work across key applications.
A new initiative aims to equip teachers with practical AI skills through hands-on workshops.
A new project from Thinking Machines, called Inkling, is now available on Hugging Face, marking a notable addition to the platform.
OpenAI's new academy lessons position ChatGPT Work as a workflow tool tailored separately to sales reps and data scientists.
OpenAI's new bio bug bounty invites external researchers to probe GPT-5.5 for realistic biological misuse risks.
A new OpenAI analysis argues that the popular SWE-Bench Pro benchmark may not reliably measure how well AI models solve real-world coding tasks.
NVIDIA highlights how French organizations are moving from AI commitments to deployed agentic systems across industries.
A new report from Anthropic renews the debate over whether freely downloadable AI models are a security risk or a necessary check on concentrated power.
A new legal AI model from Z.ai reportedly triples its base model's performance on a rigorous legal benchmark while keeping test costs low.
A new tutorial shows how to benchmark a deterministic 3-tiered Graph-RAG system against standard vector RAG, with hallucination as the key metric.