Thinking about how organisations perform in an AI-enabled world

AyEye Today · 31 October 2026

The web can now say “no training”. The terms may still say yes.

Abstract painting of blue and saffron signal lines converging before dense off-centre oxblood and black blocks.
Original artwork: Where the Terms Begin. Chiappe × OpenAI.

A common training-preference signal reduces configuration work. It cannot by itself make refusal enforceable or commercially survivable.

The story, brieflyEvidence · Interpretation

The argument

Separate declaration from enforceability. A shared token makes a path-specific preference legible; contracts, bargaining power, audit and remedy decide whether it has force.

What is new

OpenAI will honour the proposed robots.txt token. The CMA has separately opened an investigation into Meta terms that it says may deny publishers a choice over website-content use.

Why it matters

The draft separates crawling from generative training, yet permits private agreements to override the public preference.

LecXie’s practical interpretation

Publish interpretations, preserve retrieval-time evidence, audit dataset outcomes and make refusal possible without surrendering essential access.

Evidence & limits

The token is work in progress, OpenAI’s implementation is future-facing, and the CMA case is an investigation with no finding of infringement.