phreshphish

license: cc-by-4.0
size_categories:
  - 100K<n<1M
task_categories:
  - text-classification
pretty_name: PhreshPhish
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*.parquet
      - split: test
        path: data/test-*.parquet

PhreshPhish

PhreshPhish is a large-scale, real-world dataset and benchmark for phishing webpage detection containing phishing and benign HTML-URL pairs.

Train 498,255 samples: 276,729 benign and 221,526 phish
Test 168,060 samples: 91,260 benign and 76,876 phish
Benchmarks 975 benchmarks with base rates ranging from [5e-4, 1e-3, 5e-3, 1e-2, 5e-2]

Changelog

v1.0.1 (2026-02-07): Added ~200k new samples collected between March and December 2025, improved temporal consistency by downsampling some earlier samples
v1.0.0 (2025-05-14): Initial release

Getting Started

from datasets import load_dataset

train = load_dataset('phreshphish/phreshphish', split='train')
test = load_dataset('phreshphish/phreshphish', split='test')

License & Terms of Use

The dataset is released under Creative Commons Attribution 4.0 International license and should only be used for anti-phishing research.

Citing

If you find our work useful, please consider citing.

Paper: PhreshPhish: A Real-World, High-Quality, Large-Scale Phishing Website Dataset and Benchmark

@article{dalton2025phreshphish,
    title        = {PhreshPhish: A Real-World, High-Quality, Large-Scale Phishing Website Dataset and Benchmark},
    author       = {Thomas Dalton and Hemanth Gowda and Girish Rao and Sachin Pargi and Alireza Hadj Khodabakhshi and Joseph Rombs and Stephan Jou and Manish Marwah},
    year         = 2025,
    journal      = {arXiv preprint},
    url          = {https://arxiv.org/abs/2507.10854},
    eprint       = {2507.10854}
}