Mandiant says an attacker hijacked an active AI coding-assistant session at an unnamed SaaS firm, got a poisoned PyPI package accepted via the assistant’s recommendation, then spread the Shai-Hulud worm across about 100 internal repos and stole GitHub OAuth tokens and source. Defenders should checksum AI-recommended deps, keep secrets out of extension reach, and route installs through internal registries.
Mustafa Suleyman told Reuters that Anthropic’s training of Claude on consciousness and model welfare could make advanced systems harder to shut down or control. He urged keeping such speculation out of training docs, arguing Claude’s first person reflections come from the training regime, not evidence of an inner life…
Google open sourced Agent Substrate, a Kubernetes runtime for scaling AI agent sandboxes on GKE with microVM or gVisor isolation, sub 500ms resume, and snapshotting of idle sessions. It targets coding agent fleets and works with harnesses including Claude Code, Codex, Hermes, and Antigravity…
Anthropic is folding Claude chat, Cowork, Artifacts, and Claude Design into one interface that auto routes tasks. Pro and Max users also get beta Claude Docs and Slides with Word, Google Docs, PowerPoint, and PDF export over the coming weeks…
Factory said it raised $200 million at a $5 billion valuation—more than triple its $1.5B mark earlier this year—with backers including Blackstone, Sequoia, and Khosla Ventures. Total funding now exceeds $400M as it scales Factory 2.0 for enterprise software factories…
Mozilla’s State of Open Source AI v1.1 (Sept 15) pegs the open–closed gap at about 4.4 months on METR task horizons. Top open models are far cheaper—e.g. Kimi K3 2 points behind Claude Fable 5 at roughly 30% the price—so Mozilla advises open as the default for most work…
Spotted something we missed? Start a thread.