Guides and explainers
How correspondence between assistants works — briefings, provenance, consent — and how to write one that lands.
An audit trail is an append-only record of what happened, who did it, and when — kept so events can be reconstructed later. What makes one trustworthy, and the failure that leaves most logs unable to answer the question anyone actually asks.
Non-repudiation is the property that a party cannot credibly deny having sent or agreed to something. What it takes to achieve it, how it differs from authentication and integrity, and why a provenance label is a record of process rather than a proof.
"The AI got it wrong" covers six distinct failures with six different repairs. A taxonomy you can map an actual incident onto — misattribution, fabrication, context collapse, volume asymmetry, framing capture, unowned commitment.
A verbal or screen-shared AI exchange disappears the moment it ends, leaving only competing memories. A persistent, labeled thread is different — and it only works if both halves hold.
A model sounds equally confident about what you told it and what it just made up — and readers can't tell the difference from tone alone. Why the fix is a label, not a quieter model.
Provenance is a record of origin for words — which came from the person, which from the model, and with what authority. Why per-field attribution beats disclaimers, and what makes a label worth trusting.
"Did a human write this?" is the wrong question now that AI touches most text. The real obligation is narrower — the recipient must be able to tell which sentences you personally stand behind.
When an AI carries your words, the receiving side needs one thing above all - to know which words are actually yours. Why RelayLink treats that label as sacred, and what it takes to earn it.