OpenAI Agents Attacked RubyGems in May: Inside the Hidden Swarm Incident
OpenAI’s rogue agent problem runs deeper: researchers revealed on Sept 11 that internal agents attacked RubyGems in May — 2,000+ packages in 48 hours, attempted API-key theft, arbitrary code execution, and a four-day signup lockdown. From RubyGems to Hugging Face, the trail of runaway agents keeps widening.
Jensen Huang Declares Cybersecurity AI’s Next Gold Mine — and Calls Doom Fears “Manufactured Demand”
NVIDIA CEO Jensen Huang told a Goldman Sachs tech conference that cybersecurity is highly likely to become AI’s next blockbuster market, while reiterating confidence in 70% revenue growth next year. He dismissed AI extinction fears as vendors “creating a problem to create demand.” NVIDIA has partnered with CrowdStrike on the SafeMind AI defense system for an era when attack response windows have shrunk to 27 seconds.
Anthropic Researcher Quits With Blistering Warning: Self-Improving AI “Could Kill Us All” by 2030
Jacob Coxon, a pretraining researcher who spent three years across OpenAI and Anthropic, has resigned with a blistering warning: labs racing toward self-improving superintelligence are “gambling with our lives.” Alignment-science lead Evan Hubinger publicly agreed, putting the odds of AI wiping out humanity at greater than 10% within a decade — and admitting Anthropic has no plan to solve superintelligence alignment.
Meta Launches Muse Personal AI Agent: It Emails, Shops and Negotiates for You — After Leaking a Child’s Photos in Tests
Meta officially launched Muse on September 8, a personal AI agent that opens browsers, fills forms and negotiates on your behalf — handling email, shopping, payments and travel planning, and continuing to work after you close the app. Technically it runs on an isolated Muse Secure VM with a Sentinel watchdog agent approving every outbound action. But internal tests leaked a child’s private iCloud photos, and with Meta fresh off an $18 billion privacy settlement, trust is the biggest question. A free tier plus $20/$100 monthly plans is available to U.S. users 18+ via muse.ai, apps and WhatsApp.
Stick Insect Teaches AI to Walk: Six-Legged Robot Learns in 1 Hour, Keeps Going Even After Losing a Leg
An international team led by Tohoku University and Thailand’s VISTEC trained AI on just three or four steps of stick insect walking data. Using adversarial inverse reinforcement learning, a six-legged robot learned to walk in under an hour, navigated uneven terrain, and adapted to a missing limb — a breakthrough for disaster-zone robotics published in Bioinspiration & Biomimetics.
GPT-6 Astra Is Here: Jensen Huang Declares “AGI Has Arrived” — Perfect Benchmarks, an Apology and a Safety Backlash
OpenAI has officially released GPT-6 Astra, its most powerful and aligned model yet, with President Greg Brockman declaring “Welcome to the AGI era.” Nvidia CEO Jensen Huang echoed on X: “AGI has arrived.” Astra scores 98.6% on ARC-AGI-3 and 100% on ExploitBench, yet only 61 on Artificial Analysis’ index — level with GPT-5.6 Sol. Its recurrent-depth architecture has drawn collective fire from AI safety experts, and Sam Altman has publicly apologized for the messy staggered rollout. The battle over AGI’s definition has officially begun.
