← All studies

Superintelligence in Domain Names

A descriptive study of domain name registration patterns around announcements about superintelligence.

Design and scope

Generell AI chose the domain name lists before examining their registration dates. They were selected to investigate vocabulary associated with machine cognition and superintelligence. The 61 everyday nouns were chosen for similar length and commercial appeal and provide a comparison list. They are not a random or quantitatively matched sample of the domain market.

The observation is a domain name’s recorded creation date, not a person, purchase price, website or instance of language use. The main comparison uses 19–26 June 2024, inclusive, around the 19 June SSI announcement. We also examine shorter and longer intervals to describe sensitivity to the date boundaries. The September sample is a separate description, not a comparison of the strength of two responses.

Registration data were collected on 23 September 2026 using the .si and .com registries. The four CSVs contain the domain names, registration classifications and creation dates used in the analysis.

Data collection

We queried 178 domain names through the registries’ RDAP services, from 2026-09-23T20:48:03+00:00 to 2026-09-23T20:50:31+00:00.

• 162 queries returned registration records with creation dates.
• 16 queries returned “record not found”. This does not guarantee that a name can be registered.

The query log records each source URL, retrieval time, HTTP status, date-validation outcome and response hash. The collection summary also records hashes of the input files. Full registry responses are held in the research archive; the public log contains retrieval metadata and hashes. Registry access is documented by Register.si.

The creation dates in the dataset agree with the RDAP responses at the recorded precision: calendar dates for .si, and UTC hours for .com. Registry records describe the registration state at collection; they do not establish past availability or ownership history.

The June 2024 pattern

The ratio of these sample proportions is 15.95. It is not an estimate that a randomly selected cognition-related name was sixteen times more likely to be registered. All 126 .si names are marked registered in the dataset, but that does not tell us which were available before this window.

The 17 dates fall on six calendar days; eight are on 24 June. The only comparison-list date in the window is table.si, on 25 June. The full records appear below.

SSI’s founding statement puts safe superintelligence at the centre of its mission. The timing of this cluster is compatible with interest around its June 2024 launch. These records alone cannot distinguish that explanation from one buyer assembling a portfolio or other reasons for registering the names.

How the counts depend on the date window

This exploratory sensitivity analysis compares the main eight-day interval with adjacent, shorter and longer intervals. It describes how counts depend on date boundaries; the rows are not independent replications. Every row uses the same lists and denominators: 65 cognition-related names and 61 comparison names.

The dates remain concentrated near late June when the endpoint changes. Excluding 19 June leaves 16 cognition-related dates and one comparison date, so the pattern does not rely on assuming that a launch-day registration occurred after the announcement. These checks do not resolve the limits of the vocabulary selection or shared ownership.

What vocabulary does this measure?

Most names in the main list describe cognition or computing broadly. They do not contain the word superintelligence. A search of the domain labels finds only two .si labels containing superintelligen: superintelligence.si, dated July 2023, and superintelligent.si, dated June 2024. Only one lies in the main interval.

This distinction limits the title’s interpretation: the study concerns domain names associated with the subject, not a measured surge in registrations containing the exact term. The separate 13-name .com list includes three dates in 19–26 June 2024 and names dating back to 1999. It cannot establish when the word entered commercial use.

The September 2026 records

The 39-name list contains five creation dates on 22–23 September 2026, 18 earlier dates and 16 entries marked available. superintelligencepolicy.com supplies one additional date in the separate 13-name list, giving six observations across the two lists.

The .com timestamps are recorded to the hour in UTC. Relative to the reported speech start of 13:55 UTC on 22 September, all six fall within the following 24 hours. This timing comparison does not establish that the registrations were a response to the speech. The calendar-date counts above do not depend on its reported start time.

The 16 names labelled “available” in the CSV returned “record not found” in registry queries. This label does not guarantee availability for registration. The 2024 and 2026 lists differ in extension, vocabulary and observation period, so we do not rank the size, strength or durability of a response between them.

What the study supports

The selected cognition-related list contains more creation dates in late June 2024 than the selected everyday-noun list. The counts are reproducible from the dataset, and the sensitivity analysis makes their dependence on the interval explicit. The September files contain six dates close in calendar time to the announcement, across two separate lists.

We do not infer the number of independent buyers, the reasons for registration, an announcement’s causal effect or a change in industry terminology. The samples were selected before dates were examined, but they are not representative samples of the market, and individual domain records may share a buyer.

Statistical interpretation

We report counts and sample proportions without a claim of statistical significance. Independence between domain records has not been established: one buyer can register multiple names, and the comparison list does not establish the historical pool of names available to register.

An optional calculation is available through python3 analyze.py --fisher, using the table [[17, 48], [1, 60]] and the two-sided Fisher convention. A correct calculation does not establish that its inferential assumptions fit this study.

Data and reproduction

Download the four CSVs and analyze.py into one directory and run python3 analyze.py. The script uses only the Python standard library and reports counts, date windows and the literal-word subset. Download analysis.json and run python3 analyze.py --check to compare the published calculations with your local run.

• si-domains.csv — 65 cognition-related names.
• control-domains.csv — 61 everyday-noun comparison names.
• com-domains.csv — 13 superintelligence-related names.
• reaction-domains.csv — 39 September candidate names.

The query log provides registry URLs for reproducing the collection. Save retrieval times, complete responses and errors. Timeouts, rate limits and other service errors are inconclusive; a “not found” response does not guarantee availability. Later queries may return different results as records change.

Remaining limits

Coverage. The lists were selected in advance, but they cover particular words and extensions. There is no quantitative matching for historical demand, local uses of .si, price or pre-window availability. The findings apply to these lists.

Independent decisions. Registry records are not a count of independent buyers. Present-day registrant references, registrars or nameservers would not by themselves establish distinct decision-makers at the time of registration.

History and timing. Current creation dates can omit earlier registrations that were dropped. The date-only .si files cannot order launch-day events; the .com files record hours without minutes or seconds. These data do not reconstruct ownership history.

Causation. This design has no observation of motivation and no matched counterfactual for either announcement. A causal claim would require additional evidence about purchasers and competing explanations, not simply more domains or a smaller p-value.

Full domain lists

Related article: Trump’s AI adviser co-founded a firm that backs a superintelligence startup — investor connections and a summary of this study’s findings.