OpenAI has resumed frontier model training with no live internet access, replacing live browsing with a cached offline world after a September security incident.
Big Tech Moves
Read more →
The latest AI news, newest first — updated around the clock by our editors.
OpenAI has resumed frontier model training with no live internet access, replacing live browsing with a cached offline world after a September security incident.
Moonshot AI has reduced the cyber-offensive capabilities of its new Kimi K3.1 Pro model ahead of publishing its open weights, citing growing concern over AI-assisted hacking.
Widgets and Web has launched fully AI-operated investor relations websites for public companies at a fixed monthly price, with live market data and automated SEC filings.
Axon has unveiled Body 5, a body-worn camera with 25 times the AI processing power of earlier models, plus a Pro version and real-time Guardian safety alerts.
Daytona raised $60 million at a $1 billion valuation to build an Agent Cloud that gives every AI agent its own sandboxed computer.
Anthropic has open-sourced sandbox-runtime, the OS-level sandbox behind Claude Code, letting developers fence in AI agents without containers.
AWS and UC Berkeley's Infinity runtime suspends AI agents during tool calls, letting 75,000 agents idle on just 8 GB of memory — built for serverless scale.
Google Cloud's Agent Compute Engine assigns each agent session its own hardware-isolated microVM with zero credentials inside — a new security model for agentic work.
Anthropic has donated its Agent Skills standard to the Linux Foundation's Agentic AI Foundation, giving the fast-growing agent format a neutral, vendor-free home.
The UK announced £600M in AI security funding, anchored by a new AI Biosecurity Lab with up to £200M in compute and model credits from OpenAI and Anthropic.
