presentofai/ai-timeline
AI Timeline Dataset An open, dated, source-linked record of what artificial intelligence actually did between July 2025 and today. Every row is a single real-world event with a primary source attached. 3,101 events · 796 distinct publishers · 2025-07-01 to 2026-09-26 Maintained by Present of AI, a daily AI news site. Updated as the timeline grows. Why this exists Most AI datasets are benchmarks or model outputs. This one is a record of events: deployments, funding… See the full description on the dataset page: https://huggingface.co/datasets/presentofai/ai-timeline.
AI Timeline Dataset
An open, dated, source-linked record of what artificial intelligence actually did between July 2025 and today. Every row is a single real-world event with a primary source attached.
3,101 events · 796 distinct publishers · 2025-07-01 to 2026-09-26
Maintained by Present of AI, a daily AI news site. Updated as the timeline grows.
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
Schema
Realm distribution
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}
}Related
- 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