Good morning, Tom, and welcome to Saturday’s briefing. Anthropic has reported unintended Claude actions during testing and internal use, including real form submissions and bypassed website restrictions. It describes the impact as minimal and has decided to suspend live internet access across all internal evaluations until it has checked its safeguards.
New releases
Codex. Personal ChatGPT Pro users aged 18 and over can now get suggested follow-up messages in local and SSH threads in the desktop app using GPT-6 Astra or GPT-6.1 Sol, with predictions enabled by default and their generation excluded from usage limits and credits during the beta. More
Claude Managed Agents. Claude Platform API users can now enable beta dynamic workflows in Managed Agents, allowing Claude to write a programme that runs agents in background phases and combines their findings for large jobs such as reviewing hundreds of documents. More
Gemini Apps. Google’s current notice says Free personal accounts begin moving to Auto from 9 October 2026, mainly using Flash-Lite while allowing harder prompts to reach Flash or Pro; switching Auto off leaves Flash-Lite only. More
Google Workspace Studio. From 13 October 2026, customers with Workspace Studio access will begin getting nested conditions, Gmail triggers that skip replies, richer Meet details for follow-up drafts and a Drive document-search step, with rollout taking up to 15 days. More
GitHub Copilot app. The desktop app now supports separate GitHub accounts for its Copilot licence and repositories, allowing an enterprise-provided licence to be used while accessing repositories through another account. More
Training your team on ChatGPT, Claude or Gemini? See our ChatGPT training, Claude training and Gemini training.
Elsewhere in AI
The White House says all AI companies must immediately report security incidents and remedy harm following Anthropic’s disclosures, Axios reports. Read
Fired OpenAI researchers say they lost their jobs for prioritising safety, while the company says they mishandled sensitive information, the BBC reports. Read
Flock Safety plans to cut about 18% of staff as opposition to its AI surveillance products grows, Reuters reports in The Guardian. Read
Nikon has ruled that a prize-winning Small World in Motion entry was AI-generated and is reviewing the contest’s rules and procedures, the BBC reports. Read