Anthropic co-founder Jack Clark told the BBC labs may eventually need a verifiable way to shut AI off if it becomes too dangerous. UNSW’s Toby Walsh called kill switches a “complete distraction” that could invite cyberattacks, arguing for independent oversight instead.
OpenAI acquired Glass Imaging for over $300M (WSJ via TechCrunch, Sept 14). The Los Altos startup—founded by ex Apple Portrait Mode engineers—uses neural nets tuned to each phone camera to improve shots at capture time, fitting OpenAI’s rumored device plans after the $6.5B io deal…
Anthropic’s Managed Agents API now documents a third permission policy, auto : the server judges each tool call from the tool, args, and session so far, then allows, denies, or pauses for approval. It isn’t the default—set {"type":"auto"} on a toolset or per tool config…
Anthropic’s Sept 14 engineering post says Claude now authors 80% of its code and CI jobs rose 25x in six months. After three short lived patches, they redesigned test impact analysis around a journaled in memory store so listener workers stay horizontally scalable…
OpenRouter’s US In Region Routing is generally available for Business/Enterprise: requests to us.openrouter.ai stay decrypted and served only in the US (or 404). Open weight models are 60% of its US tokens, mostly Chinese (DeepSeek V4 Pro, Kimi K3, GLM 5.2 via US hosts)…
The Agentic AI Foundation (Sept 14) opened the Model Context Protocol Associate (MCPA), a 120 minute Linux Foundation exam on MCP architecture, interactions, and security—aligned to the 2026 07 28 spec. It’s the first official MCP credential as Tier 1 SDK downloads near half a billion monthly…
Spotted something we missed? Start a thread.
