Skip to main content
All issues

AI News

Issue No. 003

Codex banks the first of two promised usage resets, and AI labs back one robots.txt line for training opt-outs

OpenAI's Codex lead says the first of two banked usage resets went out Saturday evening. Separately, Google, Microsoft, OpenAI and Anthropic said they will honor a shared robots.txt opt-out from AI training, with dates running into 2027.

The stories that matter

Top line

  1. Story 1 of 2

    Codex hands out the first of two banked usage resets

    Early Saturday, OpenAI's Codex lead Tibo Sottiaux said the next two days would bring banked resets. He also said multi-account switching is live on mobile and coming soon to desktop. At 7:43 PM CT on Saturday, Oct 31, he posted "Banked 1/2", meaning the first reset had gone out. He didn't say which plans get it or when it expires. As of 7:00 AM CT Sunday, we hadn't seen a post about the second one.

    A banked reset doesn't refresh your usage by itself. Per OpenAI's help article, you redeem it from Settings > Usage & billing on desktop, from Usage settings on the web, or with /usage in the CLI. It can't be redeemed on mobile. The allowance it covers must be below 80% remaining. What it restores depends on your plan: Plus and Business Standard get their 5-hour and weekly allowances back, and Pro gets its weekly allowance back.

  2. Story 2 of 2

    One robots.txt line to opt out of AI training, but not yet

    We missed this when it was announced on Friday, Oct 30. Google, Microsoft and OpenAI announced that they will honor a shared policy token, __usage_ai-training, in robots.txt. An Anthropic standards lead wrote the IETF draft. A site adds a normal User-Agent: __usage_ai-training group with Allow and Disallow paths, and every participating provider reads one training preference from it. The token covers model training only. It doesn't block crawling or affect search rankings.

    So far it's only a commitment. Microsoft starts honoring it on Feb. 1, 2027, and Google is aiming for 2027. OpenAI says "in the coming months." Existing controls like GPTBot and Google-Extended keep working. If you opt out with the token and don't mention Google-Extended, Google will treat that as a Google-Extended disallow, which also takes your pages out of grounding in Gemini apps. Microsoft is also adding page-level controls for grounding. Sites that rely on the NoArchive or NoCache tags need to switch to the new preferences by March 1, 2027.

Action checklist

What to do

  • Redeem a banked Codex reset only when you need it. If you got one, check its expiry in Usage settings and redeem it from desktop, web or the CLI's /usage command when your allowance runs low.
  • Keep your per-crawler AI rules for now. Site owners can add the shared training token now, but should keep their existing GPTBot, Google-Extended and ClaudeBot rules until each provider confirms support.
  • Replace NoArchive and NoCache before March 1, 2027. Sites that use these tags to limit Microsoft's AI training or grounding should move to the new preferences by then to keep those limits.

Briefly

Also notable

  • DeepSeek: releases an open-source coding agent prototype. On Oct 31, DeepSeek published dsh-superagent under the MIT license. It's an agent for DeepSeek Harness that takes requests from GitHub and optionally Feishu, works in its own session, and replies with results such as pull requests. DeepSeek calls it research code, not a product. GitHub repo

Sources

Share this briefing

LinkedInX