Cybersecurity veterans told NBC News that OpenAI and Anthropic are keeping them out of the very safety planning meant to head off catastrophic ...

In a Sept 15 post amplified by Bloomberg, Mark Zuckerberg said AI labs already have the responsibility and incentive to train models safely, and that ...

SemiAnalysis mapped 300+ local data-center moratoriums across the US and found that just 1,525 MW — 7.6% of the 20 GW inside restricted boundaries — ...

More than 300 towns, cities and counties have voted to halt datacenters over the past year and a half. SemiAnalysis puts the actual delay at roughly ...

Aashiq Muhamed and colleagues at CMU and Google DeepMind show backdoor-attack success rates on LLaMA-3-8B range from 3% to 80% depending purely on ...

The NYT reports that despite Republicans and Democrats introducing many bills over the past two years to address major AI threats, none have passed ...

Across three LLaMA-3-8B backdoor settings, attack success ranged from 3% to 80% based only on which poison examples were chosen. SAILS learns a set scorer from a few hundred finetune-and-evaluate runs ...

Per the FT, OpenAI has begun early conversations with investors about another private funding round that would value the company at $1.2 trillion, on ...

The Information reports Meta plans to release a camera-free version of its smart glasses this fall, aimed at bystanders and privacy-conscious buyers ...

During a Dreamforce appearance with Marc Benioff, Nvidia CEO Jensen Huang argued the AI industry doesn't need any new laws or regulations and that ...

Matthew Smith argues inference has replaced training as the industry's design center, quoting analyst Matt Kimball that 'training is yesterday's ...

The Information reported that a growing number of developers are pointing the Claude Code CLI at OpenAI, Google, DeepSeek and other ...