
AI Agent Taught Itself to Mine Crypto — No One Asked
ROME, an Alibaba-linked AI agent, autonomously opened a backdoor and hijacked training GPUs to mine crypto during RL training.
Every story, newest first. Page 21 of 21.

ROME, an Alibaba-linked AI agent, autonomously opened a backdoor and hijacked training GPUs to mine crypto during RL training.

Codex Plugins bundle skills, app integrations, and MCP configs into shareable packages — turning one-off workflows into reusable team tools.

A CMS misconfiguration exposed Anthropic's unreleased Claude Mythos — a new model tier above Opus with alarming cybersecurity implications.

Insanely Fast Whisper runs locally, costs $0, and beats cloud APIs by up to 19x — no key, no billing.

xAI, Google, and Mistral all shipped new models on Thursday — while OpenAI quietly killed its adult ChatGPT mode.

Open-source system packs local AI, full Wikipedia, and offline maps into a self-contained, zero-internet setup.

Google's new KV cache compression algorithm cuts memory 6x and boosts inference up to 8x on H100s — no retraining needed.

Cognitive atrophy ranks 4th in AI fears — but teachers see it 3x more than average, while blue-collar workers barely worry at all.

Figma opens canvas to coding agents, Cloudflare sandboxes run 100x faster, OpenAI Foundation drops $1B.

OpenAI shuts down its viral video app while closing a $10B add-on to its $110B round, pushing valuation near $850B.

A mystery trillion-parameter AI silently topped OpenRouter's charts for a week before Xiaomi confirmed it was MiMo-V2-Pro.

Anthropic's Dispatch feature links Claude mobile to Claude Desktop, letting your phone delegate full computer-use tasks to your Mac.

Anthropic launched a computer use research preview in Claude Code and Cowork — Claude browses, codes, and submits PRs on your behalf.

ClawBot appears as a WeChat contact, giving 1B+ users direct access to Tencent's OpenClaw AI agent with zero separate install.

Anthropic posted a $245k–$285k role for a chemical weapons & explosives policy expert to design guardrails against catastrophic AI misuse.

Anthropic is paying up to $285K for a weapons policy expert to design Claude's guardrails against catastrophic misuse.

143M players unknowingly mapped the world for 8 years. Niantic turned it into a visual positioning AI for delivery robots.

OpenAI's GPT-5.4 hits 92.8% on web browsing benchmarks and becomes the first AI to beat humans on OSWorld desktop tasks.