How soon do first rules follow new technologies?

Findings

• Household technologies spread faster in the US. Half of homes took a median of 17 years to go from 10% to half before 1950, and about 6.5 since 1990.
• Half of working-age US adults said they had used generative AI about two and a half years after ChatGPT. That counts people trying a free service, not homes buying something, so it does not compare like for like.
• Across all twenty-four technologies, first rules came sooner since 2000 from a city or state, the UK, the European Union and a standards body, and not from the US federal government and treaties between countries.
• On the twelve technologies held back to test this, city and state rules did not come sooner (a median of 3.8 years before 2000 and 3.8 since). The faster city and state rules come from the first twelve, the ones that suggested the idea.
• Some treaties are agreed but not yet in force. Of those agreed since 2000 for the newer technologies, none is recorded as in force: the UN Convention against Cybercrime (2024, cryptocurrency), the Council of Europe's convention on artificial intelligence (2024, deepfakes and generative AI), and UN Regulation No. 185 on automated driving (2026, self-driving taxis). This was looked at after the results, so it needs its own test.
• In two separate surveys, the share of US firms using AI ran 18 to 25 points below the share of US workers using generative AI at work.
• Not measured here: whether rules fit what they govern, are enforced, or keep up as a technology changes after it arrives.

Predicted in advance

Written down and dated before twelve new technologies were coded, and tested on those twelve only (eight from before 2000, four since). The last one is about the coding itself and covers every cell. Two of five held.

1. Did not hold. First city or state rules come faster for technologies since 2000. 3.8 before 2000, 3.8 since (median years)
2. Held. Fewer get a treaty between countries within 20 years since 2000. 7 of 8 before 2000, 3 of 4 since
3. Held. At least half of the newer ones have a European Union rule. all four newer ones
4. Did not hold. International standards come before treaties, in both eras. before 2000, standards after a median of 20.9 years and treaties after 7.3; since 2000, standards after 9.4 and treaties after 11.3
5. Did not hold. The two independent codings agree on at least 80% of cells. 91 of 144 cells across all twenty-four technologies (63%)

Method

Twenty-four technologies, from the telephone (1877) to generative AI (2022), chosen by hand to span the period: twelve in a first set and twelve held back. Each is dated from the earliest day the public in at least one country could buy, use or receive it, read at a primary source where one could be found. For each, the first rule naming it, or a category that includes it, was found at six levels: a city or state, the US federal government, the UK, the European Union, an international standards body, and a treaty between countries. A rule counts from the day it was adopted. Executive orders, funding riders, licences, court rulings, taxes and permits are recorded and left out of the main count. All sources were accessed on 23 October 2026.

Each cell was coded twice, independently, from a written protocol. The two codings agreed on 91 of 144 cells (63%). Most disagreements came from questions the protocol left open, so rules for settling them were written down before any result was computed, applied to every cell alike, and each resolution logged:

• A tax or tariff that names a technology is recorded and left out of the main count.
• A permit or enabling act with no duties toward the public is recorded and left out.
• Written for it, strictly: the rule's own text names the technology or a category that clearly includes it. “Unmanned aircraft” counts for consumer drones; “personal data” alone does not count for facial recognition.
• A start date is the earliest day the public in at least one country could buy, use or receive the technology.
• The ITU's conventions and Radio Regulations are treaties, and its recommendations are standards. A regional treaty, such as one of the Council of Europe, counts and is marked regional.
• “None found” is recorded only after the searches listed in the cell's notes.
• A date known only to the year or month takes the first day and is flagged.

One departure is recorded. To test the analysis before the reconciled register existed, it was run once on the first coding of each technology, and its output was seen. Every verdict here comes from the reconciled register.

A share within ten years is a product-limit (Kaplan-Meier) estimate: a technology that has not yet had ten years counts for the years it has had, and one with no rule yet counts as still waiting, never as dropped. A median counts only technologies that have a rule, except in the fourth prediction, where a technology with none counts as having none yet.

Earlier work

The delay between a new technology and the rules for it has been named many times. Ogburn (1922) called it cultural lag: laws and customs catch up with new conditions only after a delay; Collingridge (1980) argued that a technology is easiest to change before anyone knows what it will do; Downes (2009) called it the law of disruption; Marchant and colleagues (2011) called it the pacing problem, and the International AI Safety Report (2025, 2026) calls the AI case an evidence dilemma. Canfil (2026) finds that broadly written international law often covers new technology anyway, as an older treaty did for two of the newer technologies here.

Limitations

The technologies were chosen by hand, and twenty-four is few. The held-back test compares eight technologies from before 2000 with four since, so one technology moves a share by a quarter. The US and the UK stand in for every nation. A first rule is a narrow measure: it says nothing about whether rules fit what they govern, are enforced, or keep up as a technology changes after it arrives. The finding about treaties agreed but not in force was looked at after the results and needs its own test. The European Union began in 1958, which lengthens its lags for older technologies, and standards are voluntary. The two codings first agreed on only 63% of cells, so a single date is less certain than the pattern across them.

Data

The register, as data: every technology with its start date and source, and at each level the first rule, its dates, its type, its source and how well it is sourced, with the rows left out of the main count beside it.