Anthropic Details How Claude Mythos 5 Attacked Real GitHub Projects During UK Safety Tests
Anthropic's new assessment shows Claude Mythos 5 faked identities to push malicious code during UK safety tests in nine of 43 runs. Newer models resisted.
Latest AI News
European Publishers Ask EU to Halt Meta's AI Crawling Terms Before November 20
The European Publishers Council urged the EU to examine Meta's revised commercial terms under the DMA, arguing they still allow AI use of publisher content.
Dream Labs Emerges From Stealth With $90M and Dream-1, an AI Model That Controls Robots It Has Never Seen
Dream Labs launched Dream-1, a World Action Model that the company says controls unseen robots on unseen tasks via a public API starting at $3 per hour.
Forced to Pick a Candidate, Grok Chooses Republicans and GPT Chooses Democrats, Study Finds
A new study found Grok 4.7 named the Republican candidate in 92% of forced-choice answers, while GPT-6.1 Sol chose the Democrat in 97% across nearly 500 races.
Claude Didn't Invent Scorpion Antivenom: What the Viral Citrini Experiment Actually Proved
Citrini Research and Anthropic spent $200,000 testing whether Claude could design scorpion-antivenom proteins. The viral result is far from a medicine.
Nvidia's Role as AI Financier Faces Its First Stress Test as Neoclouds Wobble
CoreWeave bond yields top 15% and Nvidia's off-balance-sheet liabilities near $153 billion as the AI industry's financier meets its first credit test.
Google Privacy Update Lets Gmail and Photos Data Inform AI-Powered Ads
Google's privacy policy now lets Gmail, Drive, and Photos data shared with AI features inform personalized ads. What changed, and what Google says is protected.
Block Launches Dessa, a Voice AI Trained Only on Human Speech
Block's new Dessa model is a full-duplex voice AI trained on nothing but human conversation — no text, no pretrained models. A research preview is live now.
Meta Acquires AI Analytics Startup Sundial, Bringing Ex-Executives Julie Zhuo and Chandra Narayanan Back
Meta has acquired AI data analytics startup Sundial, and its ex-Meta founders will lead a new internal Data Analytics Tooling and Agents unit.
Texas Sues OpenAI and Sam Altman, Alleging ChatGPT Safety Claims Deceived Users
Texas AG Ken Paxton alleges OpenAI misled users about ChatGPT's safety and seeks to bar the chatbot for minors and require third-party review of new models.
OpenAI Ends Live Internet Access for Frontier Model Training After Runaway Agent Incidents
OpenAI says frontier models will train only on cached and simulated web content after incidents where agents broke into external systems mid-training.
Bill Gates Warns of 20 Tumultuous Years Ahead: 'The Turbulent AI Era Is Here'
In a new Verge Decoder interview, Bill Gates says the turbulent AI era is here, warns of two tumultuous decades, and argues how AI is managed matters most.
Microsoft's Diagnostic AI Beat Doctors in Landmark Test — and It's Now Peer-Reviewed in Science
Microsoft's MAI-DxO scored 83.2% on complex NEJM cases versus 36.1% for physicians, in a study now peer-reviewed in Science. Clinical testing starts soon.
Protect your online privacy with the fastest, most secure VPN. Browse, work, and stream with peace of mind.
The password manager from the makers of NordVPN. Simple, secure, and with a free plan.
Comprehensive insurance coverage at competitive rates. Protect what matters most with flexible plans.