A History of Privacy  ·  antiquity – 2026

Who Writes
the Terms?

For most of history, privacy was a wall, a seal or a law someone else enforced. Online, it became a policy the other side wrote and you clicked to accept. A new IEEE standard, 7012 — MyTerms — turns that around: you propose the terms; if the site or service agrees, both keep the record. This is the long road to that reversal, and why it matters now that AI agents act for us.

Who has set the terms of privacy: the wall and the seal, the court and the statute, the organisation's policy, the person's software sending a signal — and, last, the person's own agreement.

Jeremy Fritzhand By Jeremy Fritzhand Open Rights Archive  ·  September 2026
Start the narrative Jump to IEEE 7012 Why it matters for AI agents

Before the Story  ·  What Privacy Is

A Short History of an Idea

Privacy is older than the word for it, and the word has meant different things in different places. The small numbers in the text are sources: every sentence is backed by the note beside it.

The Ideas  ·  In Their Authors' Words

One Word, Many Claims

Privacy has meant a closed door, a kept confidence, the right to be left alone, control over what is known about you, and the expectation that information flows only in ways that fit its context. Each idea below was put into words by someone at a particular moment, and each is quoted from the source.

Who Writes the Terms — Four Voices of the Digital Era

Online, every exchange of personal data runs on terms. The question this history keeps returning to is who proposes them, who merely accepts, and who keeps the record.

The Long Road  ·  World Map

Moments That Moved the Line

Scroll the timeline and the map follows the argument: a physician's oath, a law of windows, a house that is a castle, a sealed letter, a camera that could be carried, a data bank no one had agreed to, the first data protection law, a cookie, a signal nobody had to honour, a regulation that did have teeth — and a standard that lets a person propose the terms.

Antiquity · WALLS & SEALS
Scroll to begin the timeline →
Mapped moments by this date: —
Regions: —

Diffusion & Time

The Whole History on One Axis

Every mapped event on this site, placed by region and date. The time axis is compressed: the two thousand years before the printing press share the left tenth, and everything before 1880 the left fifth, so that the last century and a half — when privacy became a matter of machines, records and networks — can be read at all.

Turning Points, by Region

Each dot is one cited event from the map above, coloured by era. The pattern is the argument: scattered ethical, religious and legal rules for two thousand years, then from 1604 a run of court rulings and statutes in Europe and North America, and after 1970 a spread of data-protection law around the world.

Dots are this site's mapped events, each carrying its own source (hover for it). The count is of cited turning points, not of every privacy law or case.

Era Durations — Years in Each Era

The first era spans two millennia; each digital era lasts a decade or two, and the current one has run only since 2023. Log scale.

Era boundaries are editorial (DERIVED): each opens at a cited event named in the tooltip.

How Firm Is the Evidence, Era by Era

Every source attached to a mapped event, counted by verification status.

Counts of source objects on this site's mapped events, not a measure of how much is known about each era.

Interactive  ·  Year Scrubber

Scrub the History of Privacy

Drag the slider and the map fills in: each cited moment appears in its year. The cards give the count of countries with data-protection laws and of US states with comprehensive privacy laws where a cited series reaches that year, name the year each figure comes from, and say so plainly when no series reaches that far back.

— Era: —

Compressed time: the track is linear between its break years — 1600, 1880, 1970, 2000 and 2016 — not between every pair of labelled years. Colours are the eight eras.

—
Mapped moments
—
Regions with a mapped moment
—
—
Countries with Data-Privacy Laws

The dashed line is the slider year.

Mapped Moments — Cumulative

Data Protection  ·  1970 – 1993

The Law Catches Up with the Computer

The first data-protection law was passed in the German state of Hesse in 1970. By 1981, principles written in Washington, Paris and Strasbourg had given the world a shared vocabulary for records about people. They are addressed to the organisation that holds the data — though the first of them already said the person should help set the terms.

Countries with Data-Privacy Laws

The Principles, Side by Side

The fair information practice principles were an early attempt to write privacy as rules for record-keepers. Quoted from each text.

Around the World  ·  1970 – 2026

How Privacy Law Spread

Data-protection law began in Europe, but it did not stay there. Drag the year: each dot is a country's first general data-privacy law (for the United States and Canada, one covering only the federal public sector), coloured by the decade it was passed. The list beside the map names each law and its source.

—
YearCountry & lawSource

One row per country: the first general data-protection law, not sectoral rules or constitutional clauses. This site's list is a cited sample, not a census; the count of all countries with such laws is the series above.

First Laws by Decade and Region

Counted from the list above (DERIVED).

The International Instruments

Treaties, regulations and frameworks that carried privacy across borders, in order of adoption.

InstrumentBodyYearForceReach

Different Ways to Govern the Same Data

The main models countries have followed. None of them starts from terms the person proffers — which is what IEEE 7012, a voluntary standard rather than a law, sets out to do.

Hackers, Cypherpunks & Advocates

The People Who Argued It Out Online

Much of the modern argument about privacy was made first on mailing lists, at hacker congresses, in code and in leaked documents — by cryptographers, hacker clubs, activists, researchers, whistleblowers and journalists, several of them known best by a handle. Each card quotes the person's own words.

Machine-Readable Privacy  ·  1998 – 2026

Machine-Readable — but Whose Terms?

People have tried to make privacy machine-readable for a quarter of a century. Most attempts let the organisation state its policy, or let the person's browser send a preference the organisation could ignore. The table follows each one to its fate, and asks the same four questions of each.

Mechanism Who proffers the terms Machine-readable Legal force Record kept by Fate

Each row cites its specification or the body that published it.

Surveillance & Reckoning  ·  2013 – 2022

The Surveillance Economy, Counted

By 2013 the scale of collection had become measurable — by journalists, regulators and researchers — and law began to answer it with fines, rights and, for the first time, signals a business must honour.

GDPR Fines, Cumulative

GDPR Fines per Survey Year

US States with Comprehensive Consumer Privacy Laws

The Human Cost

When Records Were Turned on People

Data protection was not invented in the abstract. Registers, censuses, files and profiles compiled for one purpose have repeatedly been used for another — to deport, to intern, to watch, to manipulate. Each row states what the cited source records.

Place & year What happened Scale Consequence Source

Figures are what the cited source states for that place and date. Where a figure is contested or partial, the note says so.

The Destination  ·  IEEE 7012-2025

MyTerms: The Person as First Party

IEEE 7012 is a standard for machine-readable personal privacy terms. Instead of a site presenting its policy and asking you to accept, you — or software acting for you — proffer one of a small set of standard agreements kept in a public roster; the site's software accepts it, declines or offers one alternative from the same roster, and once they agree, each side keeps an identical record.

The Principles Built into the Agreements

The Launch Agreements

Two relationship agreements for ongoing service, and three for one-off contributions of data. MyTerms says each comes in plain language, in machine-readable form and as a legal text — the three readings the standard requires.

The Agentic Turn  ·  2023 – today

Why It Matters Now: Terms for the Age of Agents

AI agents now browse, buy, book and fill in forms for people. Each of those errands meets someone else's terms. An agent cannot meaningfully read a cookie banner, and a person cannot review every agreement an agent makes on their behalf. What an agent needs is its principal's terms in a form it can present, and a record that the other side agreed to them.

One Errand, Step by Step
Illustrative — not a specification

How an agent could carry its principal's terms into a transaction. Each step says what in the published MyTerms and IEEE 7012 material supports it, and says so plainly where the public material does not specify the detail.

    What Could Still Go Wrong

    The honest limits. Two earlier machine-readable efforts failed for reasons that still apply.

    Jurisdiction Matrix

    Who Honours a Machine-Readable No?

    A person's terms need somewhere to stand. Some privacy laws already oblige businesses to honour an automated objection or an opt-out preference signal; most say nothing about it. Each cell is cited.

    Jurisdiction Main law Maximum penalty Automated signals AI rules

    "Automated signals" asks whether the law requires a business to recognise a person's automated or machine-readable choice or objection (for example GDPR Article 21(5), or opt-out preference signals under the California and Colorado rules). It is a narrow question, not a ranking of privacy regimes.

    Corrections

    Claims That Did Not Survive Checking

    Privacy history is full of good stories. These are the popular versions the research pass could not confirm, with what the record shows instead.

    Open Archival Imagery

    Pictures & Primary Visual Sources

    Rights-cleared images, every one public domain, CC0 or openly licensed. Each card names its creator, license and record; the thumbnails are mirrored in the repository with their provenance, and each opens the full-size original.

    Source Audit

    Every Claim, Its Institution, Its Status

    Loading citation audit…

    Institution Data points Confirmed Pending Derived Access Link

    CONFIRMED = the cited page states the claim, and the supporting sentence was checked against it. PENDING = the right institution or a credible secondary source, not yet read in the primary record. DERIVED = editorial or computed, with its inputs named in the note. Hunt list: ARCHIVAL_RESEARCH_PROMPT_PRIVACY.md · Rights: CONTENT_LICENSE.md

    About

    What This Is, and How to Read It

    An open-data history of privacy, told as a history of who gets to set the terms.

    Privacy is old; the idea that a person could state their own privacy terms to an organisation, in a form a machine can read and, its backers say, a court can enforce, is new. This project follows the line between the two — with a world map, a set of charts, a year scrubber, a lineage of machine-readable privacy and a rights-cleared picture archive — and it shows its working for every claim on the page.

    It is a domain fork of Indigo: The Blue That Dyed the World and Downtown Tampa: same architecture, same citation standard, a different subject. The site is static: no build step for the reader, no backend, no cookies and no analytics. Everything it knows lives in one cited data file, compiled from the research notes in the repository.

    How the evidence is graded

    Every claim carries a source object naming the institution, the document, the date and a status, and most carry the exact sentence from the source that supports them. The research pass fetched each cited page it could reach and checked the quoted sentence against it (the few it could not are marked Pending); a second, independent pass tried to refute every record.

    • Confirmed The cited page states exactly this, and the quote was checked against the page. Hero and footer numbers must be confirmed.
    • Pending The source is the right one or a credible secondary account, but the claim has not yet been read in the primary record. Treat as provisional.
    • Derived Computed or editorial: an era boundary, a cumulative count, an illustrative scenario. The note names the inputs. Do not cite these as primary figures.

    The Source Audit prints the live tally. Where sources disagree, the disagreement is printed, not averaged.

    A note on the destination

    This study argues that IEEE 7012 matters. It tries to earn that argument rather than assume it: the case for the standard in the age of agents is set out claim by claim with evidence, the scenario is labelled as an illustration, and the limits — adoption, enforcement, and the fate of the machine-readable efforts that came before — have their own section.

    Reuse

    The compiled dataset is released under CC BY 4.0. Embedded images are public domain, CC0 or openly licensed and credited on each card; anything whose reuse terms are unclear is linked rather than copied. See CONTENT_LICENSE.md for the full policy, and open an issue if you hold rights in something here.