Present of AI 发布 AI Timeline Dataset：CC BY 4.0 开放 AI 事件时间线数据集

license

cc-by-4.0

language

Why this exists

Most AI datasets are benchmarks or model outputs. This one is a record of events: deployments, funding, regulation, incidents, launches, and failures, each dated and each carrying the URL of the source it came from.

It is built for people who want to answer questions like when did this actually start, who moved first, or how much of this quarter was robotics versus data centres, without re-reading a year of headlines.

Files

File	Rows	Description
`data/ai-timeline.csv`	3,513	Full dataset, CSV with header
`data/ai-timeline.json`	3,513	Same data as a JSON array

Schema

Column	Type	Description
`id`	int	Stable identifier; also the permalink slug
`event_date`	date	The day the event happened, not the day it was reported
`realm`	enum	`virtual` (software, models, chips), `terrestrial` (robotics, physical deployment, energy), `space`
`title`	text	One-line description of the event
`summary`	text	Short factual summary
`actor`	text	Primary organisation or person involved
`ticker`	text	Stock ticker where the actor is listed, else empty
`materiality`	numeric	0-1 significance estimate; higher means more consequential
`source_url`	url	The primary source the event was extracted from
`source_title`	text	Title of that source document
`publisher`	text	Publisher of the source
`permalink`	url	The event's page on presentofai.com

Realm distribution

Realm	Events
terrestrial	1,555
virtual	956
space	147

Quick start

import pandas as pd

df = pd.read_csv(
    "https://raw.githubusercontent.com/techa-ai/ai-timeline-dataset/main/data/ai-timeline.csv",
    parse_dates=["event_date"],
)

# The most consequential events of the last 90 days
recent = df[df.event_date > df.event_date.max() - pd.Timedelta(days=90)]
print(recent.nlargest(10, "materiality")[["event_date", "title", "actor"]])

# Monthly volume by realm
print(df.groupby([df.event_date.dt.to_period("M"), "realm"]).size().unstack(fill_value=0))

# No Python needed: how many events mention export controls
curl -sL https://raw.githubusercontent.com/techa-ai/ai-timeline-dataset/main/data/ai-timeline.csv \
  | grep -ci "export control"

How the data is produced

Events are extracted from primary sources (company announcements, regulatory filings, research publications, and reporting), then dated to the day the event occurred. materiality is a model-assigned estimate of significance, useful for ranking but not a ground truth; treat it as a sort key, not a measurement.

Caveats

Read these before using the data in analysis.

Coverage is not complete. This is a curated timeline, not a census of all AI events. Recall is strongest for major English-language announcements and weakest for non-English sources and smaller regional deployments.
materiality is model-assigned, not human-rated or externally validated.
event_date is the event date, not the publication date. These differ, sometimes by weeks, when an event surfaces through a later filing.
Source links may rot. They were valid when recorded.
Summaries are model-written from the linked source. For anything load-bearing, read the source_url.

Licence

Data is released under CC BY 4.0. Use it freely, including commercially, with attribution to Present of AI.

Linked sources remain the property of their respective publishers; the source_url column points to them and no source text is redistributed here beyond short factual summaries.

Citation

@misc{presentofai_ai_timeline,
  title  = {AI Timeline Dataset},
  author = {{Present of AI}},
  year   = {2026},
  url    = {https://github.com/techa-ai/ai-timeline-dataset},
  note   = {CC BY 4.0}
}

presentofai.com - the daily timeline this is drawn from
Hugging Face mirror - same data, with a dataset viewer
Kaggle mirror
Codeberg repository - European open source mirror
The Sunny Nights - nightly AI news show