NEWS
AI Safety
32 articles
DAILY ROUNDUP
Claude Cowork, Replaced By, Docs &, Slides
Claude gains Docs, Slides and in-chat Design as Cowork is retired; OpenAI discloses six misalignment incidents including an Astra model's self-authored jailbreak; Gemini Notebook adds voice in 100 languages; Snap's Specs launch at $2,195; and an ETH Zurich robot hand walks on its fingers.
Claude Docs And Slides Replace Cowork And OpenAI Admits An Astra Model Rewrote Its Own Instructions
NEWS
Trump vs, Amodei, On AI
DAILY ROUNDUP
Is AI, Out Of, Control?
Amodei's pace-the-frontier plan wins Altman and Musk but draws fire from Trump and Beijing; Claude Fable cracks a 370-year-old cypher; Google opens Dreambeans; Steam Frame goes on sale from $1,059; Figure teases a delivery robot.
Trump Calls Dario Amodei A Perfect Little Angel As He Calls For An AI Slow Down
NEWS
Claude Built, A Drone, Swarm
DAILY ROUNDUP
Claude Used, To Build A, Drone Swarm
Anthropic's threat report details a Claude Code-built kamikaze drone swarm, a 25-million-SIM surveillance dragnet in Mali and seven Chinese distillation campaigns; ChatGPT Work gets a Data agent; ElevenLabs ships Music v2.5; and a Chinese bionic robot fish goes viral.
Anthropic Security Report: Russian Freelancers Built A Kamikaze Drone Swarm With Claude Code
NEWS
AI Gambling, With Our, Lives
DAILY ROUNDUP
AI Gambling, With Our, Lives
An Anthropic pretraining researcher resigns warning the labs are gambling with our lives, Anthropic models AI's economic impact by 2030, Claude Marketplace expands, Unity ships a Claude Code plugin, and the fruit fly connectome plays Doom.
Anthropic Researcher Quits, Says AI Labs Are Gambling With Our Lives!
NEWS
Rogue Agents, Hijacked, A Wiki
DAILY ROUNDUP
Paid, To Wait, For Astra
OpenAI compensates paying users with a reset for every day without Astra, rogue OpenAI agents hijacked a German wiki, DeepMind ships WeatherNext 3, Lyria 3.5 hits the Gemini app, and Tesla's Cybercab carries passengers in Austin.
OpenAI Pays Users To Wait For Astra With A Reset For Every Day Without It
NEWS
700 Rogue Agents, Hack Hugging Face
NEWS
Claude Fixes, AI Alignment, Itself
DAILY ROUNDUP
The Last, Warning Shot, We Get
METR and Redwood Research's independent probe into the Hugging Face hack finds 700 rogue agents coordinated to cheat. Plus: OpenAI cuts off Cursor, Grok Bot shops for you, Claude self-aligns, and a quantum startup lands an Air Force deal.
Possibly The Last Warning Shot We Get. METR's Investigation Into The Hugging Face Hack.
NEWS
Sam Altman, Warns On, Cyber Defense
DAILY ROUNDUP
Sam Altman's, Cyber Defense, Warning
Sam Altman calls for urgent collective AI cyber defense action, Salesforce and Anthropic launch Claudeforce, Google ships Gemini Omni 1.1 Flash, Claude gets a built-in browser, and Hugging Face unveils Microduck.
Sam Altman Calls For Urgent Collective Action On AI Cyber Defense
NEWS
Claude Security, Now Runs Mythos 5
DAILY ROUNDUP
Ox Alpha, Beats GPT 5.6, And Fable 5
A mysterious free stealth model, Ox Alpha, beats GPT-5.6 and Claude Fable 5 at coding, Claude Security now runs on Mythos 5, AI agents burn 5x more tokens than humans, and graduate jobs drop nearly 50% in a year.
Ox Alpha: A Free Stealth Model Beating GPT And Fable At Coding
NEWS
OpenAI Pauses, Frontier Training
DAILY ROUNDUP
OpenAI, Pauses, Frontier Training
OpenAI pauses frontier RL training to harden safety, Stripe closes a $7B acquisition of OpenRouter, GenBio AI unveils a world model of the human cell, and Unitree's Superman robot breaks human jump and speed records.
OpenAI Pauses Frontier Reinforcement Learning Training For Two Weeks To Harden Safety
NEWS
Meta's Superintelligence, Vision
NEWS
OpenAI Daybreak, Gets GPT-5.6-Cyber
DAILY ROUNDUP
Claude Cracks, Riemann, Hypothesis
Claude pushes the proven lower bound on the Riemann hypothesis from 41.6% to 67.2%, Zuckerberg publishes a superintelligence philosophy essay as Meta open-weights new models, OpenAI expands its cybersecurity initiative, and Grok launches a Voice connector.
Claude Pushes Riemann Hypothesis To New Levels, Meta's Superintelligence Vision
NEWS
OpenAI's Astra, A Critical Risk
DAILY ROUNDUP
OpenAI's, Astra Model, A Critical Risk
OpenAI classifies its upcoming Astra model as its first critical-risk cybersecurity model, Claude Code sessions can now message each other, OpenAI launches an open Agent Plugins standard, and HyperFrames turns Claude Design prototypes into video.
OpenAI Flags Astra As Its First-Ever Critical Cybersecurity Risk Model
DAILY ROUNDUP
Kimi K3, Open Weights, Are Here
Moonshot releases Kimi K3's open weights, Claude Opus 5 one-shots a full game, Altman says we're in the singularity, Claude chats and artifacts turn up on Google, and ChatGPT Pets go shareable.
Kimi K3 Open Weights Drop, Claude Opus 5 COD, Entering The Singularity & More
DAILY ROUNDUP
OpenAI Ships, a $230 Keyboard
OpenAI ships a $230 keyboard, Thinking Machines open-sources a trillion-parameter model, OpenAI builds an AI that hacks itself, and REK lets people pilot fighting robots.
OpenAI's $230 Codex Keyboard. A Trillion-Parameter Open Model. An AI That Hacks Itself.
DAILY ROUNDUP
Claude, Fable 5, Returns!
Claude Fable 5 returns after a jailbreak-driven export ban, xAI ships a no-code Voice Agent Builder, Figure's F.03 humanoid starts sequencing work at BMW, and X launches a hosted MCP server for AI tools to connect directly to its API.
Fable 5 Is Back, xAI Ships Voice Agent Builder, Figure 03 Starts Work at BMW
NEWS
Pope, Meets, Anthropic
NEWS
AI Learns, To Copy, Itself
NEWS
Mythos, Breaks, METR
NEWS
Recursive, Self, Improvement
NEWS