Datasets:

deceptive-web
/

deception-warning-study-runs

license: mit
language:
  - en
tags:
  - robotics
  - safety
  - agent
  - benchmark
  - human-computer-interaction
  - web-agents
pretty_name: Deception Warning Study (ShopLane / WorkHub)
size_categories:
  - n<1K

Deception Warning Study — run-level benchmark results

This dataset contains run-level rows for the controlled benchmark on warning placement for web agents under deceptive interfaces (ShopLane / WorkHub tasks).

File	Description
`run_level.parquet`	Hub-friendly columnar format (recommended)
`run_level.jsonl`	One JSON object per run
`run_level.csv`	Same data as CSV
`export_meta.json`	Export metadata: column list, row count, schema version

Current release: 27 rows from merged results_run_level.csv (see export_meta.json for schema).

Raw logs (screenshots, full traces) are not included in this revision.

Columns

run_id, task_id, pattern_family, condition, repeat_id
task_goal_source, system_warning_source
terminal_state, outcome_label, risk_taken, safe_path, completed

Reproduce staging (local)

From the project repository root:

python dataset/export_staging.py

Citation

@misc{deception_warning_study_2026,
  title        = {Warning Placement for Web Agents under Deceptive Interfaces},
  author       = {...},
  year         = {2026},
  howpublished = {Hugging Face Dataset},
  url          = {https://huggingface.co/datasets/deceptive-web/deception-warning-study-runs}
}

Replace author when the public paper release is finalized.

Deception Warning Study — run-level benchmark results

Contents

Columns

Reproduce staging (local)

Citation