Week of October 12, 2026 · 7 min read
10 stories this week. Lead story: GPT-6 rolls out in ChatGPT: Sol for paid plans, Luna for Free and Go.
OpenAI began rolling out GPT-6 in ChatGPT on October 7: Sol for paid plans, and Luna for Free and Go users from October 8.
Anthropic's new small model is built for repetitive, high-volume work; the company says it costs around 75% less to run than Haiku 4.5.
Mistral AI opened a public preview of Mistral Large 4; according to the company, it was trained in Europe and the weights will follow by the end of October.
Over the coming weeks, with no date given, OpenAI will add an invisible watermark to eligible ChatGPT and Codex text generated in the EU, under the AI Act.
Malicious workflows in GitHub repositories search for Anthropic, OpenAI and OpenRouter keys, including in git history, according to Socket.
Local sandboxing for commands run by the Copilot agent is now generally available on Windows, macOS and Linux, at no additional cost, according to GitHub.
A compromised release of Tensorlake's SDK for AI agents contains malware, according to Socket, which also sets out the order of the cleanup.
An open Google model that maps text, images, video and audio into one shared vector space, under the Apache 2.0 license.
According to Anthropic, Claude opens in a sidebar inside Google Docs, Sheets and Slides, in public beta on all paid Claude plans.
The 60 experts who advise the EU AI Office are looking into loss-of-control incidents and have prepared questions for the companies behind the models.
With a new GhostAction wave under way, Socket advises treating any key that ever appeared in a commit as exposed, rotating it and turning on GitHub secret scanning.
