Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed because of a cast error
Error code: DatasetGenerationCastError
Exception: DatasetGenerationCastError
Message: An error occurred while generating the dataset
All the data files must have the same columns, but at some point there are 2 new columns ({'value', 'statistic'}) and 4 missing columns ({'status', 'detail', 'check', 'section'}).
This happened while the csv dataset builder was generating data using
hf://datasets/prospex-ch/swiss-ai-claims-register/tables/03_capital_test.csv (at revision 458b110c1806ca09ffc75be8656906f72b969a8b), ['hf://datasets/prospex-ch/swiss-ai-claims-register@458b110c1806ca09ffc75be8656906f72b969a8b/tables/03_capital_test.csv', 'hf://datasets/prospex-ch/swiss-ai-claims-register@458b110c1806ca09ffc75be8656906f72b969a8b/tables/04_proportional_hazards_test.csv']
Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1848, in _prepare_split_single
writer.write_table(table)
~~~~~~~~~~~~~~~~~~^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 765, in write_table
self._write_table(pa_table, writer_batch_size=writer_batch_size)
~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 773, in _write_table
pa_table = table_cast(pa_table, self._schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
statistic: string
value: double
-- schema metadata --
pandas: '{"index_columns": [{"kind": "range", "name": null, "start": 0, "' + 506
to
{'section': Value('string'), 'check': Value('string'), 'status': Value('string'), 'detail': Value('string')}
because column names don't match
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1694, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1850, in _prepare_split_single
raise DatasetGenerationCastError.from_cast_error(
...<4 lines>...
)
datasets.exceptions.DatasetGenerationCastError: An error occurred while generating the dataset
All the data files must have the same columns, but at some point there are 2 new columns ({'value', 'statistic'}) and 4 missing columns ({'status', 'detail', 'check', 'section'}).
This happened while the csv dataset builder was generating data using
hf://datasets/prospex-ch/swiss-ai-claims-register/tables/03_capital_test.csv (at revision 458b110c1806ca09ffc75be8656906f72b969a8b), ['hf://datasets/prospex-ch/swiss-ai-claims-register@458b110c1806ca09ffc75be8656906f72b969a8b/tables/03_capital_test.csv', 'hf://datasets/prospex-ch/swiss-ai-claims-register@458b110c1806ca09ffc75be8656906f72b969a8b/tables/04_proportional_hazards_test.csv']
Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
section string | check string | status string | detail string |
|---|---|---|---|
row counts | rows: companies_ai_purpose = 2,860 | PASS | 2860 |
row counts | rows: companies_population = 796,113 | PASS | 796113 |
row counts | rows: registry_events = 2,042,782 | PASS | 2042782 |
row counts | rows: name_changes = 88,872 | PASS | 88872 |
row counts | rows: purpose_changes = 137,637 | PASS | 137637 |
row counts | rows: trademarks_ai = 3,634 | PASS | 3634 |
nets | strict net = 2,806 (ai_purpose file) | PASS | 2806 |
nets | ml flag = 230 | PASS | 230 |
nets | deep/generative flag = 21 | PASS | 21 |
nets | population net: strict = 2,806 | PASS | 2806 |
nets | population net: loose-only = 54 | PASS | 54 |
nets | population strict + loose-only = ai_purpose rows | PASS | 2806+54 vs 2860 |
regex parity | python net reproduces SQL flag: match_ai_strict | PASS | 0 disagreements |
regex parity | python net reproduces SQL flag: match_ml | PASS | 0 disagreements |
regex parity | python net reproduces SQL flag: match_deep_generative | PASS | 0 disagreements |
regex parity | loose net matches every exported row | PASS | 2860/2860 |
declensions | declension suffixes are load-bearing for the strict net | PASS | 858 strict hits match only with German declension suffixes |
joins | every AI company is in the population file | PASS | 0 orphans |
joins | event companies outside the population file are a minority | PASS | 236,597 of 828,876 event companies (28.5%) have no registry profile |
joins | name-change companies all appear in the event feed | PASS | 0 orphans |
joins | purpose-change companies all appear in the event feed | PASS | 0 orphans |
joins | resolved trademark owners resolve into the population | PASS | 1,630 of 3,634 trademarks linked, 0 unresolvable |
joins | population company_id is unique | PASS | null |
joins | ai_purpose company_id is unique | PASS | null |
geography | canton ZH strict count = 597 | PASS | 597 |
geography | canton VD strict count = 369 | PASS | 369 |
geography | canton ZG strict count = 369 | PASS | 369 |
geography | canton GE strict count = 289 | PASS | 289 |
geography | canton TI strict count = 223 | PASS | 223 |
geography | canton counts agree between the two exports | PASS | [] |
events | event feed starts 2018-08-29 | PASS | 2018-08-29 |
events | event feed ends on or before the export date | PASS | 2026-08-31 |
events | event count INCORPORATION = 388,710 | PASS | 388710 |
events | event count LIQUIDATION = 156,588 | PASS | 156588 |
events | event count DELETED = 245,692 | PASS | 245692 |
events | INCORPORATION coverage reaches back to the feed start | PASS | 2018-08-29 .. 2026-08-31 |
events | LIQUIDATION coverage reaches back to the feed start | PASS | 2018-08-29 .. 2026-08-31 |
events | DELETED coverage reaches back to the feed start | PASS | 2018-09-03 .. 2026-08-31 |
events | NAME_CHANGED coverage reaches back to the feed start | PASS | 2018-08-29 .. 2026-08-31 |
events | PURPOSE_CHANGED coverage reaches back to the feed start | PASS | 2018-08-29 .. 2026-08-31 |
founding dates | founded_year is sparse, as caveat 2 says | PASS | 116,229 of 796,113 (14.6%) |
founding dates | event-derived dates add coverage on top of founded_year | PASS | founded_year 116,229 -> combined 399,185 (event 330,154) |
founding dates | exact year agreement is the modal case | PASS | 36,614/47,171 (77.6%) exact, 83.5% within +/-1 year |
founding dates | clock disagreement is one-directional (event postdates founded_year) | PASS | event >= founded+2y: 7,032 (14.9%); event <= founded-2y: 771 -- likely re-registrations, not births |
spells | spells: company_id unique | PASS | 387873 |
spells | spells: all durations positive | PASS | 0.5 |
spells | spells: censored companies have no exit date | PASS | null |
spells | spells: every observed exit carries an exit type | PASS | null |
spells | spells: no exit after the censoring date | PASS | 2026-08-31 00:00:00 |
spells | spells: entry never precedes the feed start | PASS | 2018-08-29 |
trademarks | trademark name hits = 14 | PASS | 14 |
trademarks | nearly all trademark hits are in goods/services | PASS | 3624 |
trademarks | trademark owner country CH = 2,239 | PASS | 2239 |
trademarks | trademark owner country US = 559 | PASS | 559 |
trademarks | trademark owner country DE = 202 | PASS | 202 |
trademarks | trademark owner country CN = 162 | PASS | 162 |
trademarks | filing dates span 1983 to 2026 | PASS | 1983-10-14 .. 2026-09-02 |
AI claims in Swiss commercial register purpose statements, 2018–2026
Computed results and hand labels from a full-population analysis of AI claims in the Swiss commercial register. Every Swiss company declares a legal purpose (Zweck / but / scopo) that is inscribed in the register at the cost of a notary and bounds what the company may do. As of 31 August 2026, 3,310 companies carry an AI term in their purpose.
The canonical version of this dataset lives at https://prospex.ch/open-data/swiss-ai-claims-register/
This dataset accompanies the analysis published at https://prospex.ch/blog/ai-companies-in-switzerland-what-the-commercial-register-shows/
The source data is the Swiss Official Gazette of Commerce (SHAB/FOSC), the official publication of record for the Swiss commercial register. The register feed covers 29 August 2018 to 31 August 2026.
Licensed CC BY 4.0. Attribution is a condition of the licence, and a link to the canonical page above is the attribution we ask for.
Contents
tables/
65 CSV files, one per computed result. Each file is named
{notebook}_{table}.csv, where the notebook number maps to a research
question:
| Prefix | Topic |
|---|---|
00_ |
Validation checks and re-registration contamination |
01_ |
Temporality: quarterly AI share of incorporations, pre/post-ChatGPT ratios |
02_ |
Geography: canton rate ratios, denomination sensitivity, net ranking shifts |
03_ |
Morphology: legal form mix, capital distributions |
04_ |
Survival: hazard ratios, Kaplan-Meier horizons, match quality, sensitivity |
05_ |
Renaming: AI name adoptions and abandonments, timing, provenance |
06_ |
Trademarks: filing trends, owner composition, register overlap |
07_ |
Extended net: recovered companies, compound audit, undercount by language |
08_ |
Purpose amendments: amendment lag, quarterly flow, stock, withdrawals |
See DATA_DICTIONARY.md for the tables cited in the blog post.
labels/
Hand labels that a rerun cannot reproduce:
name_ai_labels.csv: 234 company-name verdicts (is this name an AI claim?).semantic_threshold_labels.csv: 90 purpose-text labels for the embedding threshold analysis.compound_net_exclusions.csv: 4 false-positive exclusions from the compound keyword net.
nets.py
The keyword regex patterns that define "an AI claim" in German, French,
Italian and English. Includes the strict net (spelled-out "artificial
intelligence"), the loose net (adding machine learning and deep learning
terms), the compound net (hyphenated German compounds like KI-gestützt),
the misspelling net, and the name net used for rename analysis.
export.sql
The SQL that produced the source data from the Prospex production database. Included for provenance: it shows which tables and columns were queried and which regex filters were applied.
Key numbers
| Measure | Value | Table |
|---|---|---|
| AI claimants (extended net) | 3,310 | tables/07_net_comparison.csv |
| AI claimants (strict net) | 2,806 | tables/07_net_comparison.csv |
| AI share of 2026 incorporations | 2.80% | tables/01_quarterly_share.csv |
| Pre/post-ChatGPT ratio (founding clock) | 4.4× | tables/01_pre_post_chatgpt.csv |
| Zug rate ratio vs national | 2.32× | tables/02_canton_rate_ratios.csv |
| Companies that added AI later | 603 (18.2%) | tables/08_amendment_vs_incorporation.csv |
| Median amendment lag | 2.2 years | tables/08_amendment_lag.csv |
| Three-year survival, AI claimants | 94.6% | tables/04_survival_at_horizons.csv |
| Hazard ratio (AI vs matched controls) | 0.98 (0.79–1.21) | tables/04_survival_results.csv |
| Renames into AI | 63 | tables/05_transitions.csv |
| Trademark-purpose overlap | 110 of 941 Swiss AI trademark owners (11.7%) | tables/06_register_trademark_overlap.csv |
Three keyword definitions
The blog post's headline count (3,310) uses the extended net, which combines three layers:
- Loose net (2,860): spelled-out "artificial intelligence" in four languages, plus machine learning, deep learning and generative AI terms.
- Compound net (+434): hyphenated German AI/KI compounds
(
KI-gestützt,KI-basiert,Al-Protokoll), including l/I confusions the register contains, minus four hand-excluded false positives. - Misspelling net (+16): misspelled variants the register actually
contains (
Künstliche Inteligenz,Artifical Intelligence).
Citing this dataset
Sidorenko, Semion (2026). AI claims in Swiss commercial register purpose statements, 2018–2026 [dataset]. Zenodo.
Related datasets
The monthly panel of Swiss commercial register publications, covering all event types and all cantons: https://huggingface.co/datasets/prospex-ch/swiss-commercial-registry-monthly-panel
- Downloads last month
- 80