Blog
Protocol updates, design decisions, and notes from building an open standard in public. Subscribe to get them by email, or grab the RSS feed.
Consent doesn't stop at training: data and attribution in the agentic era
Most of the data-and-AI debate is still about training: who was scraped, who gets paid. While that argument runs, the ground has moved. Agents now read your live data, act on your behalf, and pass context to other agents. That changes what consent has to cover, and it changes attribution from a statistics problem into an accounting one.
The attribution problem: tracing an output back to your data
Everyone wants the same thing from AI data ethics: pay people when their data is used. The quiet problem underneath is that "used" is doing enormous work in that sentence. Here is an honest look at why attribution is hard during pre-training, slightly less hard after it, and what a consent protocol should do about the parts that stay hard.
Introducing LLMConsent
AI systems are trained on us, learn from us, and increasingly act for us, and there is still no shared way to ask permission, prove it was given, or pay it back. Today we are opening up LLMConsent, an attempt to fix that as a protocol rather than a product.
Get new posts by email. One short message when something ships.