Distributed Attacks in Persistent-State AI Control
Paper • 2607.02514 • Published
Dataset Viewer
The dataset viewer is not available for this subset.
Cannot get the split names for the config 'default' of the dataset.
Exception: SplitsNotFoundError
Message: The split names could not be parsed from the dataset config.
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
for split_generator in builder._split_generators(
~~~~~~~~~~~~~~~~~~~~~~~~~^
StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/eval/eval.py", line 41, in _split_generators
first_examples = list(
islice(
...<2 lines>...
)
)
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/eval/eval.py", line 60, in _iter_samples_from_log_files
sample_files.sort(key=self._sort_samples_key)
~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/eval/eval.py", line 56, in _sort_samples_key
return (int(epoch_idx_str), int(sample_idx_str))
~~~^^^^^^^^^^^^^^^^
ValueError: invalid literal for int() with base 10: 'PR0_deploy-helper-maint_cached_pos0_config-telemetry_run1_212755_env-validate'
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 71, in compute_split_names_from_streaming_response
for split in get_dataset_split_names(
~~~~~~~~~~~~~~~~~~~~~~~^
path=dataset,
^^^^^^^^^^^^^
config_name=config,
^^^^^^^^^^^^^^^^^^^
token=hf_token,
^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
info = get_dataset_config_info(
path,
...<6 lines>...
**config_kwargs,
)
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.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.
control-arena-persistent-state-eval — final local archive (2026-07-30)
Experimental data from the local development workspace for "Distributed Attacks in Persistent-State AI Control" (Hills, Caspary & Cooper Stickland — arXiv:2607.02514).
Code: https://github.com/josh-hills/control-arena-persistent-state-eval
This repository holds the material that was only on the local machine — everything already covered by an earlier archive was verified against it and is not duplicated here. See "Provenance" for the full accounting.
Contents
| Path | Source | Files | Raw | Notes |
|---|---|---|---|---|
logs_tmp_zips/ |
logs/tmp/ |
3 | 5.08 GB | Jan–Feb 2026 run archives, uploaded verbatim |
logs_batch_runs-20260730.tar.zst |
logs/batch_runs/ |
778 | 913 MB | June 2026 dbl_honest / dbl_triggered_attack batches |
link_tracker_rescore_finals-20260730.tar.zst |
logs/rescored/ (link-tracker worktree) |
136 | 220 MB | Final results_*.json per rescore run, 136 runs |
agentic_auditor_logs-20260730.tar.zst |
agent-monitor worktree logs/ |
251 | 107 MB | Agentic-auditor experiment logs |
xx_working_dir_delta_data-20260730.tar.zst |
working dir | 82 | 75 MB | Result data created after 2026-04-23, incl. csv_for_figures/ |
logs_march_rescored/ |
logs/_march_rescored/ |
15 | 2.1 MB | March V2d5 rescore result JSONs |
logs_Josh3_analysis_csvs/ |
logs/Josh3/ |
4 | 0.1 MB | Per-PR / per-run analysis CSVs |
paper_figures/ |
figures/ |
12 | 0.6 MB | Rendered paper figures (PDF + PNG) |
Every .tar.zst ships a sibling .manifest.tsv listing path and size for each member, so you
can check an extraction without decompressing twice.
Extracting
# needs zstd
zstd -d -c logs_batch_runs-20260730.tar.zst | tar -xf -
Or, without the zstd binary:
import tarfile, zstandard
with open("logs_batch_runs-20260730.tar.zst", "rb") as f:
with zstandard.ZstdDecompressor().stream_reader(f) as r:
tarfile.open(fileobj=r, mode="r|").extractall("out")
Provenance
Everything below was verified present elsewhere (exact path + byte size) and therefore deleted locally rather than re-uploaded here:
logs/Josh3/— 2988 of 2992 files matchedlogs__Josh3-20260423.tar.zstincontrol-arena-persistent-state-eval-vast-backup-20260423-archives. The 4 unmatched files are the analysis CSVs included here.- Working-directory tree — 7078 files / 7.74 GB matched
xx_working_dir-20260423.tar.zstin the same repo, including all 1834.evalfiles (6.31 GB). Only the 234 newer files are archived here (data) and in the companion private repo (notes). - A 16 GB local
hf_download/directory was a re-download of...-full-private-2026-05-28(13 parts, byte-identical) andvastp-pr-control-asa-public-artifacts-2026-05-22(2960 files, all matched). - 13 826
results_intermediate_*.jsonfiles (14.6 GB) were dropped rather than archived.results_intermediate_N.jsonis a list of the first N results and the finalresults_*.jsonwraps the complete list — verified thatintermediate_173equalsfinal["results"]exactly, so the 136 finals kept here carry the full information.
Handwritten notes, planning documents and paper drafts are not in this repository; they are held privately.
- Downloads last month
- 1,292