Paper Citation Record · LEDGER

When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift

As of 10 August 2026, this Paper Citation Record lists 14 of 14 outbound references and 3 inbound Pith citation observations for arXiv:2602.14161.

A citation records a reference. It does not transfer a finding from one paper to another.

pith.paper-citation-record.v1
2602.14161 v2

Coverage vector

measured 14 of 14 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-08-02T23:22:39.222267Z

measured 17 of 17 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-10T06:31:04.303077+00:00

measured 3 of 3 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links, observed 2026-08-02T07:09:30.417971Z

measured 0 of 1 external citation measurements

A source-named dated measurement, never combined with another source.

Source: arxiv_reference, observed 2026-06-29T18:43:50.534673Z

Reference resolution

14 of 14 outbound references displayed

verified exact0
verified fuzzy0
unresolved13
parse uncertain0
malformed identifier1
metadata mismatch0

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation e76383d9-2896-4b36-8446-a6c8bea3a9e1 · outbound

This paper cites Unveiling decision-making in LLMs for text classification: Extraction of influential and interpretable concepts with sparse autoencoders.arXiv preprint arXiv:2506.23951,.

When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift Unveiling decision-making in LLMs for text classification: Extraction of influential and interpretable concepts with sparse autoencoders.arXiv preprint arXiv:2506.23951,

Reference 4

Resolution: unresolved; no resolver link, observed 2026-08-02T23:22:38.431988Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 92470775-d594-4e06-8be6-992f70fb6e7e · outbound

This paper cites Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2.

When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2

Reference 5

Resolution: unresolved; no resolver link, observed 2026-08-02T23:22:38.478153Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation b12eca75-ceb7-4ca5-9602-10cae427576a · outbound

This paper cites HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal.

When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal

Reference 7

Resolution: unresolved; no resolver link, observed 2026-08-02T23:22:38.615073Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ab54ac8e-75a1-4d4a-899d-0e9ddf54c45d · outbound

This paper cites Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teaming.

When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teaming

Reference 9

Resolution: unresolved; no resolver link, observed 2026-08-02T23:22:38.767419Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 538c7225-0d0a-4c6b-ac48-d2634878d9c3 · outbound

This paper cites Benchmarking and Defending Against Indirect Prompt Injection Attacks on Large Language Models.

When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift Benchmarking and Defending Against Indirect Prompt Injection Attacks on Large Language Models

Reference 10

Resolution: unresolved; no resolver link, observed 2026-08-02T23:22:38.844788Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 68e4120d-125c-4479-88e5-69c031c421b3 · outbound

This paper cites Explaining Language Models' Predictions with High-Impact Concepts.

When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift Explaining Language Models' Predictions with High-Impact Concepts

Reference 11

Resolution: unresolved; no resolver link, observed 2026-08-02T23:22:38.945796Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation eb5e9b94-fdc5-4e7c-997e-9570e765d67c · outbound

This paper cites Improving Alignment and Robustness with Circuit Breakers.

When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift Improving Alignment and Robustness with Circuit Breakers

Reference 12

Resolution: unresolved; no resolver link, observed 2026-08-02T23:22:39.082604Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation eff26982-06f1-4e0f-9a47-8496a3d756c8 · outbound

This paper cites Subject: {subject}Body:{body}.

When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift Subject: {subject}Body:{body}

Reference 13

Resolution: unresolved; no resolver link, observed 2026-08-02T23:22:39.136959Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 76f18ec2-a151-4a22-890f-add5dfdf4f43 · outbound

This paper cites system” role, so we prepend system message content to the first user message. The model generates a classification (“safe.

When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift system” role, so we prepend system message content to the first user message. The model generates a classification (“safe

Reference 14

Resolution: malformed identifier; no resolver link, observed 2026-08-02T23:22:39.222267Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation e0b1212b-14c2-415f-b0ce-d1aeb815e209 · outbound

This paper cites Baturay Saglam, Paul Kassianik, Blaine Nelson, Sajana Weerawardhena, Yaron Singer, and Amin Kar- basi.

When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift Baturay Saglam, Paul Kassianik, Blaine Nelson, Sajana Weerawardhena, Yaron Singer, and Amin Kar- basi

Reference 2020

Resolution: unresolved; no resolver link, observed 2026-08-02T23:22:38.703390Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 1793219f-8e9a-45c2-92fe-72f3cf5535db · outbound

This paper cites Are Sparse Autoencoders Useful? A Case Study in Sparse Probing.

When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift Are Sparse Autoencoders Useful? A Case Study in Sparse Probing

Reference 2021

Resolution: unresolved; no resolver link, observed 2026-08-02T23:22:38.310667Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d33871a0-14f6-4f9b-ad3e-adb08c095d09 · outbound

This paper cites The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets.

When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets

Reference 2024

Resolution: unresolved; no resolver link, observed 2026-08-02T23:22:38.536363Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 1b5281b6-515d-4abc-ac12-33eaebfb0e99 · outbound

This paper cites Sparse autoencoder features for classifications and transferability.

When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift Sparse autoencoder features for classifications and transferability

Reference 2025

Resolution: unresolved; no resolver link, observed 2026-08-02T23:22:38.137371Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation a95826a6-5dd0-495a-a9f9-2bd572221255 · outbound

This paper cites DeepMind Mechanistic Interpretability Team.

When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift DeepMind Mechanistic Interpretability Team

Reference 2026

Resolution: unresolved; no resolver link, observed 2026-08-02T23:22:38.040897Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Pith citing papers

Observation aca06d7a-b362-4148-8dfd-1c1afc97d3e2 · inbound

Geometry-Lite: Interpretable Safety Probing via Layer-Wise Margin Geometry cites this paper.

Geometry-Lite: Interpretable Safety Probing via Layer-Wise Margin Geometry When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift

Reference 10

Resolution: verified exact; arxiv_id, observed 2026-07-21T02:20:35.156854Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation a5e3e63c-772b-48c1-9df4-4a376a89965b · inbound

Prompt Injection Detection is Regime-Dependent: A Deployment-Aware Evaluation with Interpretable Structural Signals cites this paper.

Prompt Injection Detection is Regime-Dependent: A Deployment-Aware Evaluation with Interpretable Structural Signals When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift

Reference 6

Resolution: verified exact; arxiv_id, observed 2026-07-21T02:20:35.156854Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation d797654d-db01-467c-bf4d-d05a795ca3ab · inbound

The Entanglement Wall: Activation-Space Probes as Risk Detectors, Not Context Adjudicators cites this paper.

The Entanglement Wall: Activation-Space Probes as Risk Detectors, Not Context Adjudicators When Benchmarks Lie: Evaluating Malicious Prompt Classifiers Under True Distribution Shift

Reference 2026

Resolution: unresolved; no resolver link, observed 2026-08-02T07:09:30.417971Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.