The AI Arms Race: A Former DeepMind Exec Warns of Looming Disaster
Verity Harding, ex-DeepMind policy lead, warns that the US government’s nationalist push for AI dominance risks escalating into catastrophe.
All news
982 articles — page 40 of 99.
Local LLMs get smarter without taking over your inbox
New tools let local AI assistants read only what they need from your emails, avoiding full inbox access and privacy risks.
CISA flags Langflow flaw, feds must patch by Friday
U.S. government agencies face a tight deadline to fix a critical Langflow auth bypass flaw actively exploited in AI agent frameworks.
Tiny Python tool watermarks AI text to prove its origin
A new open-source library embeds invisible cryptographic signatures in LLM outputs, letting companies reliably trace generated text back to their models.
GPT-5.6 arrives after U.S. government delay
OpenAI's latest model launches Thursday after a government-imposed pause, setting new cost-performance benchmarks in AI coding tasks.
Meta’s smart glasses get a privacy safeguard against LED tampering
Meta will disable cameras in updated smart glasses if their privacy LED is tampered with, aiming to deter modders.
SpaceX enters AI race with new model this week
Elon Musk’s SpaceX is reportedly launching its first major AI model in collaboration with Cursor, marking a bold step into the competitive AI landscape.
Accenture confirms data breach after hacker claims 35 GB theft
Accenture has confirmed a security incident after a hacker offered 35 GB of its source code and data for sale. Here’s what we know and why the breach matters.
Google’s Knowledge Graph learns the hard way: markup isn’t destiny
A developer’s experiment shows why third-party signals—not clever schema—decide who gets a Google entity node and when.
Ant Group’s LingBot-Vision redefines spatial AI with boundary-focused models
Ant Group’s Robbyant releases LingBot-Vision, a 1B-parameter self-supervised Vision Transformer family that treats image boundaries as its primary training signal—boosting dense spatial tasks while cutting compute needs.