AI & ML

Alibaba Reveals It Built the Viral HappyHorse Video AI Model

Alibaba's HappyHorse-1.0 debuted anonymously, shot to the top of global video AI benchmarks, and was confirmed as Alibaba's best-performing video model to date.

AI Is Now Finding Security Holes That Survived Decades of Human Review

From Anthropic's Claude Mythos to OpenAI's Codex Security, AI agents are discovering critical vulnerabilities faster and at a scale no human team can match.

ByteDance's Seedance 2 Is Now Globally Available Through Third Party Platforms

Claude Code's New /loop Command Turns Your AI Agent Into a Background Worker

Anthropic's Claude Code now ships a native /loop command that schedules recurring prompts on a cron-based interval, giving developers a built-in autonomous monitoring layer without any external tooling.

Anthropic Built an AI Too Dangerous to Release So It Formed Project Glasswing

Anthropic's new $100M cybersecurity alliance gives vetted partners exclusive access to Claude Mythos Preview, an AI model that can find and exploit software vulnerabilities better than most humans.

AI Has Already Replaced Parts of the Job for 1 in 5 American Full-Time Workers

A new Epoch AI and Ipsos survey of 2,000 U.S. adults finds AI is actively displacing work tasks for 20% of full-time employees, with replacement outpacing new job creation.

Intel and Google Deepen Collaboration to Advance AI Infrastructure

The two tech giants are aligning their hardware and software roadmaps to challenge Nvidia's dominance in the data center.

Anthropic's Most Powerful AI Model Redefines Cybersecurity Threats

The race to build the most capable AI system just became a cybersecurity crisis waiting to happen.

Project Glasswing: Securing Critical Software for the AI Era

A new initiative aims to tackle one of the most overlooked vulnerabilities in modern AI systems by hardening the open-source dependencies that power them.

Gemma 4 Could Be the Moment Open-Source AI Stops Playing Catch-Up

Google's latest open-weight model family arrives with benchmarks that challenge proprietary systems head-on, raising serious questions about the future of closed AI development.

MIT's CompreSSM Compresses AI Models While They Train and Delivers 4x Speedups

A new MIT-led technique called CompreSSM shrinks state-space AI models during training itself, cutting compute costs without sacrificing accuracy.

Meta Released Its First Superintelligence AI Model to Replace Llama

Muse Spark handles complex reasoning and multimodal tasks but abandons the open weights approach of previous generations.