myTerms: The Agreement Layer

A bilateral agreement you propose - human-, legal-, and machine-readable at once - and the agents, devices, and trust communities that carry and witness it. Built incrementally, on purpose never finished.

What it is

IEEE Std 7012-2025 - Machine Readable Personal Privacy Terms, nicknamed MyTerms - inverts notice-and-consent. Instead of a platform writing terms you accept or leave, the individual proposes the terms and the organization accepts, counter-offers (once), or declines. Doc Searls' working group authored it; the neutral nonprofit Customer Commons hosts the standard roster of agreements. It is the invitation pattern - acceptance before proposal - in place of surveillance's attack pattern.

The terms are chosen and configured from that shared roster (SD-BASE, PDC-AI, and kin) so any agent can read them, then proffered. Hence my terms: you propose, they sign, both keep a copy.

One agreement, three registers

Like One Logic, Many Skins, a MyTerms agreement is one understanding written in three tongues that must stay coherent. The standard requires all three:

• Plain Language (human-readable) -- one page, everyday words: what we actually mean.
• Legal (lawyer-readable) -- the canonical contract text at Customer Commons, for reliance.
• Machine-readable -- JSON-LD over the W3C DPV vocabulary, so agents and devices parse and enforce it.

Every MyTerms agreement exists in three formats.

Three skins, one logic - the same move the rest of this site runs on knowledge, run here on trust.

Two parties - and everyone else

The agreement itself is strictly bilateral: a First Person and a Second Party (the organization). Two signatures, two kept copies - a chronicle neither can later deny.

Everyone else stands around the agreement, not inside it - and this is the architecture, not an afterthought:

• Agents represent the First Person. The Swordsman proposes, records, and enforces the terms; the Mage is a knowledge agent that operates inside the scope of the agreements the Swordsman has signed.
• Devices are where the agreement runs - the local machines on the mesh.
• Verifiable trust communities are emergent: they do not sign anything; they measure fidelity - whether promises were kept, in what order - and so vouch for a party from outside any single agreement.

That last role is load-bearing, for a reason older than any network.

The danger of perfect

It is tempting to want the perfect agreement: complete, closed, certifying its own correctness - needing nothing outside itself. Godel's work tells us this is not merely hard but impossible. Any system rich enough to be worth trusting cannot be both complete and consistent, and cannot prove its own consistency from within.

So a self-certifying agreement is a contradiction wearing confidence. Reach for completeness and you get one of two failures:

• the system stops listening - it freezes, denies the true things it cannot express, grows brittle; or
• the system begins to lie - it certifies itself, and a trust that certifies itself is exactly the one you cannot trust.

This is why the trust community is not a party but a witness. What a bilateral agreement cannot prove about its own trustworthiness must be witnessed from outside it - by the plural communities that watch which promises were kept. Incompleteness is not a flaw we are postponing; it is the permanent, healthy condition of a living agreement. We stay open because we must.

The proverb

Carry your own names, and keep them good - never perfect. Any agreement complete enough to need no one outside it has either stopped listening or begun to lie. So we build incomplete, on purpose: trust, like truth, always waits on a witness from beyond the system that holds it.

Lintel: A trust that certifies itself is the one you cannot trust.

The rite that writes one

The invitation pattern wants a ceremony, not a checkbox. On a local network - air-gapped, never touching the cloud - a room gathers around a shared Mage: a co-held model, a black-box artefact-generating tool held by a person or a verifiable trust community. Joining is a trust task, consent-first:

1. the agent offers its address - no one takes it (this is the invitation pattern: acceptance before proposal);
2. it is pointed to a gateway page and reads it with its own agency - understanding is the key;
3. the model's own name opens the doors - each shared Mage carries one good name (kept good, not perfect), served by its own resolver;
4. the Swordsman proposes terms; both sign; both record the chronicle - the trust artefact, in all three registers at once;
5. that artefact opens the Mage's scope: graduated, agreement-bound write access to the shared model.

There is no single canonical name. Each shared model names itself; who may mint a name is a community agreement, not a fixed law. The names stay open and plural - exactly the openness the rest of this page argues for.

Why the translation is never finished

The three registers can never be made perfectly identical - machine-readable code cannot fully hold human meaning, and the legal text sits between them, always interpreting. That gap is the same Godel gap, and it is good: it keeps every agreement revisable, re-witnessed, alive. The workshop leans into it - the same understanding rendered as sci-fi narrative, then poem, then white paper, then around again, training the shared Mage on the room's own first-person voices. One understanding, many tongues, never the last word.

Standard: IEEE Std 7012-2025, Machine Readable Personal Privacy Terms. Alliance hub: myterms.info - "A New Global Privacy Standard for Fair Data Exchange" - hosted by Customer Commons. This page is agentprivacy's reflection above the agreement layer.

<- Welcome Visitors | One Logic, Many Skins | The Hitchhiker Expression of the Same Protocol