CoDNA Request access

The CoDNA briefing · 2 November 2026 · ქართულად

What changed in Claude and ChatGPT in October 2026

This is the first CoDNA briefing. Once a month, it covers what changed in Claude and ChatGPT that a company should know about, and what to do about it. October was a busy month: OpenAI set a date for retiring custom GPTs, both vendors changed their usage rules, Claude moved into Google Docs, and ChatGPT got a new default model.

Three things to do before December

1. Move your custom GPTs to plugins before 11 December. OpenAI is retiring custom GPTs on every plan. After 11 December they and their pages stop working. In each GPT, My GPTs → Migrate to plugin turns its instructions into a skill and its files into reference files. Two things don't carry over: custom actions and the GPT's chosen model. A GPT that calls your own systems through custom actions needs that connection rebuilt, for example as a custom MCP server. Make a list of the GPTs your people rely on, migrate them, and test each one. (OpenAI)

2. Check what you use AI for against the new usage rules. Anthropic's updated policy takes effect on 12 November. Where Claude affects someone's health, legal rights, finances, livelihood or access to essential services, a qualified person must be able to review and change its recommendations, and the people affected must be told AI was used. That covers uses such as screening job applicants or assessing loans. OpenAI's updated policies take effect on 20 November. Among other changes, a service that gives others direct access to OpenAI's models, such as a model router or AI gateway, now needs OpenAI's approval. (Anthropic, OpenAI)

3. Engineers: check what's pinned to old models and settings. On 14 October, GPT-5.5 left Codex for everyone who signs in with ChatGPT; GPT-6.1 Sol replaces it on paid plans. OpenAI asks you to update scripts, scheduled tasks and workspace defaults that still choose gpt-5.5. If you push Claude Code settings to your engineers, note one change in version 2.1.305: an MCP server your allowlist used to filter out may now load. The changelog says to block such servers with deniedMcpServers instead. Since version 2.1.316 (31 October), settings you manage centrally for Team and Enterprise logins apply without each person approving them first. And fast mode for Claude Opus 4.8 ends on 30 November, in the API and in Claude Code; move anything that uses it to Opus 5.5. (OpenAI, Claude Code, Claude API)

Claude

  • Claude in Google Docs, Sheets and Slides. A Claude sidebar now edits inside the file, and by default asks before each edit. Separate connectors let Claude create and edit Google files from a chat, following your existing sharing permissions. It's a beta on all paid plans. It needs turning on in two places: a Google Workspace admin deploys the add-on, and on Team and Enterprise a Claude owner enables the connectors. (Anthropic)
  • API credits in the Team plan. Since 7 October, a Team plan includes API credits every month: $20 per Standard seat and $100 per Premium seat, pooled, up to $500. They're for internal tools built on the API; interactive Claude Code isn't covered. To claim them, an owner links one Claude Console organisation, and only support can change that link later, so choose it carefully. Unused credits don't roll over. (Anthropic)
  • Code Review is ready for general use on Team and Enterprise. Once an admin installs the GitHub app and picks repositories, Claude reviews every new pull request, usually in 6 to 10 minutes. You set how deep it goes for each repository: Quick, Standard or Deep. That setting is the main control on cost. (Anthropic)
  • Dashboards and Motion. Dashboards are live charts built from your data platforms or connected apps. Motion makes short animated explainers you can download as video. Claude Docs, Slides and Design are out of beta. (Anthropic)
  • Plugins that change Claude Code. Plugins for Claude Code can now carry "mods", code that changes how it behaves and runs with the same access to the machine. On Team and Enterprise, owners choose which plugin marketplaces are allowed. Now is a good time to set that list. (Anthropic)
  • Claude Haiku 5.5, a new small model, is cheaper and faster, and suits high-volume automations built on the API. Code written for Haiku 4.5 may need changes before switching, because some settings it accepted now return errors. (Anthropic)
  • Skills now belong to a cross-vendor standard. Anthropic handed the Agent Skills format to the Agentic AI Foundation, part of the Linux Foundation, where engineers from Anthropic, Google, OpenAI and others maintain it. A skill you write for your company isn't tied to one vendor. (Anthropic)

ChatGPT

  • GPT-6 is now the model in Chat for Business and Enterprise. It switched on 7 October, for everyone at once. Its answers can mix text with charts and interactive parts. In Work and Codex, OpenAI added GPT-6.1 Terra on 26 October, a cheaper model for everyday tasks that makes a seat's allowance go further. OpenAI's release notes list it for Business and Enterprise. (OpenAI)
  • Plugins are where you manage what ChatGPT can do. Plugins and their marketplaces are now in the Admin Console, and admins can switch up to 25 on or off at once. Your own systems connect as a custom MCP server added as a plugin, and on Business only owners and admins can publish one to the workspace. On 21 October, Business settings moved into the Admin Console too, so update any internal how-to guides. (OpenAI)
  • API credits in Business, too. Since 28 October, eligible Business workspaces can claim API credits every month, much like Claude's Team plan: $20 per paid Standard seat and $100 per Premium seat, pooled, up to $500. They're for apps and agents built on the API; Codex used with an API key isn't covered. The person who claims them must own both the workspace and the API organisation that receives them, and each workspace links to one organisation. Unused credits don't roll over. On 29 October OpenAI also raised the Work and Codex usage included with Business Premium seats. (OpenAI, release notes)
  • ChatGPT for Word is on by default since 1 October, on Business and Enterprise. Use of it comes out of the same pool as Codex and other premium features. If you don't want it, a workspace admin can turn it off, and your Microsoft 365 admin decides whether the add-in is allowed at all. (OpenAI)
  • Audio uploads. People can now upload recordings for transcripts and summaries. If meeting recordings shouldn't go into ChatGPT, say so in your AI policy. (OpenAI)
  • Data residency is reaching Business. OpenAI says it is rolling out gradually: you choose where content is stored when you buy. Georgia isn't among the regions, and with a region outside the US a copy of each prompt and answer is still kept in the US for a limited time, for abuse monitoring. (OpenAI)
  • Also new: Slides you can create and edit together, and locked chats, which people can hide behind a PIN. Locked chats don't change how long chats are kept. (OpenAI)

On an Enterprise plan

  • Claude:
    • Quick questions and longer tasks now share one conversation, and tasks that need code or files run in Anthropic's cloud. Before it reaches your company, set spend limits and decide on "Efficient mode", which keeps people on the cheaper models.
    • Claude Docs, Slides and Design have been on by default since 15 October.
    • Inference hooks are ready for general use. They let your own security system allow or block each prompt before Claude sees it. (Anthropic, release notes)
  • ChatGPT:
    • People who must move out of a personal ChatGPT account on your domain now get exactly seven days to export their chats, so tell them before you add them.
    • The old admin-managed sync for Google Drive and SharePoint is gone. ChatGPT now reaches them live, under each person's own permissions.
    • Usage reports were reorganised, with a new per-person export.
    • Since 28 October, ChatGPT no longer moves from Instant to a higher thinking level on its own. People choose the level in the model picker when a task needs more reasoning.
    • Since 26 October, most Enterprise workspaces can't create new custom GPTs; use Plugin Creator instead. Workspaces with an approved deferral keep their GPTs until 11 February 2027. (OpenAI, FAQ)

Dates to keep

  • 12 November: Anthropic's new usage policy takes effect.
  • 20 November: OpenAI's new usage policies take effect.
  • 30 November: fast mode for Claude Opus 4.8 ends.
  • 11 December: custom GPTs retire.
  • 14 December: the standalone Claude Design site closes; move design systems into Claude before then.

Setting one of them up for your company? Our free guide, Setting up Claude or ChatGPT for your company, covers plans, sign-in, data rules and the first tasks to automate.

Questions, or something you'd like covered next month? Reply to this email. It comes straight to me.

Tornike Razmadze, CoDNA

Past issues