Paper Citation Record · LEDGER

LLM Safety From Within: Detecting Harmful Content with Internal Representations

As of 8 August 2026, this Paper Citation Record lists 80 of 80 outbound references and 3 inbound Pith citation observations for arXiv:2604.18519.

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

pith.paper-citation-record.v1
2604.18519 v1

Coverage vector

measured 80 of 80 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-05-10T04:33:54.058475Z

measured 83 of 83 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-07T06:34:17.273281+00:00

measured 3 of 3 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links, observed 2026-06-30T21:28:25.577739Z

measured 0 of 1 external citation measurements

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

Source: pith, observed 2026-06-30T21:35:04.761308Z

Reference resolution

80 of 80 outbound references displayed

verified exact12
verified fuzzy36
unresolved1
parse uncertain1
malformed identifier0
metadata mismatch30

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation d5b2a56b-da7c-4d7e-99ac-83fb41c9db39 · outbound

This paper cites ToxicChat: Unveiling Hidden Challenges of Toxicity Detection in Real-World User-AI Conversation.

LLM Safety From Within: Detecting Harmful Content with Internal Representations ToxicChat: Unveiling Hidden Challenges of Toxicity Detection in Real-World User-AI Conversation

Reference 1

Resolution: metadata mismatch; arxiv_id, observed 2026-05-10T12:20:23.136237Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation ca5b829a-b70c-420a-ab51-3c147be53ba6 · outbound

This paper cites Proceedings of the AAAI conference on artificial intelligence , volume=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Proceedings of the AAAI conference on artificial intelligence , volume=

Reference 2

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:36:04.338928Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 5d61a91f-2d50-47b3-a9f9-22c53dd92841 · outbound

This paper cites AEGIS: Online Adaptive AI Content Safety Moderation with Ensemble of LLM Experts.

LLM Safety From Within: Detecting Harmful Content with Internal Representations AEGIS: Online Adaptive AI Content Safety Moderation with Ensemble of LLM Experts

Reference 3

Resolution: metadata mismatch; arxiv_id, observed 2026-05-10T12:20:23.127272Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation ef8cd7b4-e8c6-4fcc-a861-7dcac8db4b44 · outbound

This paper cites Aegis2.0: A Diverse AI Safety Dataset and Risks Taxonomy for Alignment of LLM Guardrails.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Aegis2.0: A Diverse AI Safety Dataset and Risks Taxonomy for Alignment of LLM Guardrails

Reference 4

Resolution: verified exact; arxiv_id, observed 2026-05-10T12:20:23.119498Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 3e4d273c-0bd7-403f-bafe-ec9e1246d42a · outbound

This paper cites SimpleSafetyTests: a Test Suite for Identifying Critical Safety Risks in Large Language Models.

LLM Safety From Within: Detecting Harmful Content with Internal Representations SimpleSafetyTests: a Test Suite for Identifying Critical Safety Risks in Large Language Models

Reference 5

Resolution: verified exact; arxiv_id, observed 2026-05-10T12:20:23.111469Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 3f57f7dc-df1c-4f30-927d-2128b3986f66 · outbound

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

LLM Safety From Within: Detecting Harmful Content with Internal Representations HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal

Reference 6

Resolution: metadata mismatch; arxiv_id, observed 2026-05-11T04:06:33.939395Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation d00c611a-d5db-4180-95cc-80bc559621db · outbound

This paper cites Advances in Neural Information Processing Systems , volume=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Advances in Neural Information Processing Systems , volume=

Reference 7

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:36:04.331178Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 1116b0a4-0e5d-4143-9f40-f09da8eb3e57 · outbound

This paper cites PKU-SafeRLHF: Towards Multi-Level Safety Alignment for LLMs with Human Preference.

LLM Safety From Within: Detecting Harmful Content with Internal Representations PKU-SafeRLHF: Towards Multi-Level Safety Alignment for LLMs with Human Preference

Reference 8

Resolution: metadata mismatch; arxiv_id, observed 2026-05-10T12:20:23.122827Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 448b001b-4065-4a18-a656-900e9c4bfb9f · outbound

This paper cites Advances in Neural Information Processing Systems , volume=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Advances in Neural Information Processing Systems , volume=

Reference 9

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:36:04.335141Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 6efb3c7b-b79d-4f91-a7e1-05a7c2bb9202 · outbound

This paper cites XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models.

LLM Safety From Within: Detecting Harmful Content with Internal Representations XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models

Reference 10

Resolution: metadata mismatch; arxiv_id, observed 2026-05-15T06:51:50.894684Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation e0afad29-862c-4db9-aa1b-a349345a1af3 · outbound

This paper cites LLMs Encode Harmfulness and Refusal Separately.

LLM Safety From Within: Detecting Harmful Content with Internal Representations LLMs Encode Harmfulness and Refusal Separately

Reference 11

Resolution: metadata mismatch; arxiv_id, observed 2026-07-07T03:18:08.851986Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation ea42c941-7525-4501-9450-11bffdbb0f0b · outbound

This paper cites Advances in Neural Information Processing Systems , volume=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Advances in Neural Information Processing Systems , volume=

Reference 12

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:36:04.319477Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 7f1df501-f39a-4bbc-937e-38143a422378 · outbound

This paper cites Qwen3Guard Technical Report.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Qwen3Guard Technical Report

Reference 13

Resolution: metadata mismatch; arxiv_id, observed 2026-05-13T22:33:37.835217Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 84a353e3-dee7-49ab-9b6a-83160348045d · outbound

This paper cites Jailbreak Attacks and Defenses Against Large Language Models: A Survey.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Jailbreak Attacks and Defenses Against Large Language Models: A Survey

Reference 14

Resolution: metadata mismatch; arxiv_id, observed 2026-05-15T02:20:44.922788Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 7645e6d1-bd18-4654-b0f6-c8e614faa764 · outbound

This paper cites Defending Large Language Models Against Jailbreak Attacks via Layer-specific Editing.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Defending Large Language Models Against Jailbreak Attacks via Layer-specific Editing

Reference 15

Resolution: metadata mismatch; arxiv_id, observed 2026-05-10T12:20:23.143551Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation cc88ce6c-433a-4c59-9e53-59accdf3f005 · outbound

This paper cites Advances in neural information processing systems , volume=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Advances in neural information processing systems , volume=

Reference 16

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:36:04.323469Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 60ae356d-b6f3-4147-ae6d-213293efd60f · outbound

This paper cites Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining , pages=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining , pages=

Reference 17

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:36:04.327399Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 523891dd-59c8-4857-a9b1-cfa5e10863a8 · outbound

This paper cites Findings of the Association for Computational Linguistics: ACL 2024 , pages=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Findings of the Association for Computational Linguistics: ACL 2024 , pages=

Reference 18

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:36.978394Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 7ab56c69-546a-4139-a7be-3b6b496fbf07 · outbound

This paper cites The Eleventh International Conference on Learning Representations , year=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations The Eleventh International Conference on Learning Representations , year=

Reference 19

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:37.008478Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 1bd26b10-29c1-4ccf-a2be-4546be3cd448 · outbound

This paper cites Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing , pages=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing , pages=

Reference 20

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:37.011054Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 5a8ce3e2-c714-4813-9926-af1fd2a9d3b9 · outbound

This paper cites BERT Rediscovers the Classical NLP Pipeline.

LLM Safety From Within: Detecting Harmful Content with Internal Representations BERT Rediscovers the Classical NLP Pipeline

Reference 21

Resolution: verified exact; arxiv_id, observed 2026-05-10T12:20:23.051467Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation b18c8fae-7c8a-41f2-8f99-48f1ad45b8c4 · outbound

This paper cites Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

Reference 22

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:36.932781Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 43d98db9-d6d8-498c-bf60-daebb467c144 · outbound

This paper cites The information bottleneck method.

LLM Safety From Within: Detecting Harmful Content with Internal Representations The information bottleneck method

Reference 23

Resolution: metadata mismatch; arxiv_id, observed 2026-05-10T12:20:23.068250Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation e34d1d47-aa1f-4d77-8861-8fe651e41ff1 · outbound

This paper cites 2022 , journal=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations 2022 , journal=

Reference 24

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:37.002119Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 69f7e846-7b42-45c1-a132-6da1787dae30 · outbound

This paper cites Transactions on Machine Learning Research , year=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Transactions on Machine Learning Research , year=

Reference 25

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:36.981356Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 69121aba-0375-44d3-ba70-9283af3100a9 · outbound

This paper cites an unresolved cited work.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Unresolved cited work

Reference 26

Resolution: unresolved; raw_fallback, observed 2026-05-22T04:34:36.991231Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 42bdf367-d710-42e8-8428-dcbef79212a7 · outbound

This paper cites Proceedings of the 2022 conference on empirical methods in natural language processing , pages=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Proceedings of the 2022 conference on empirical methods in natural language processing , pages=

Reference 27

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:36.998239Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 5b69e8a9-a4ab-4deb-abb0-dd3659881546 · outbound

This paper cites Advances in Neural Information Processing Systems , volume=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Advances in Neural Information Processing Systems , volume=

Reference 28

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:36.987886Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation a4b800ca-a3da-43b0-a7b1-a1fda654f7fb · outbound

This paper cites Layer by Layer: Uncovering Hidden Representations in Language Models.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Layer by Layer: Uncovering Hidden Representations in Language Models

Reference 29

Resolution: metadata mismatch; arxiv_id, observed 2026-05-15T16:30:37.615120Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 0184305d-8bb6-4eff-83ea-c80f32d7ecf2 · outbound

This paper cites Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations

Reference 30

Resolution: metadata mismatch; arxiv_id, observed 2026-05-10T18:59:26.864775Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation f26b44fe-ecee-4a67-9ae5-c4390c5e1a6a · outbound

This paper cites Organization of Knowledge and Advanced Technologies.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Organization of Knowledge and Advanced Technologies

Reference 31

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:36.994809Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 8fec285d-6f79-4c4b-af1c-b823e1b66dcf · outbound

This paper cites Efficient LLM Moderation with Multi-Layer Latent Prototypes.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Efficient LLM Moderation with Multi-Layer Latent Prototypes

Reference 32

Resolution: verified exact; arxiv_id, observed 2026-06-02T04:04:22.908302Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation d09d643a-6e56-4864-a8ff-6dafdace5ff8 · outbound

This paper cites Safety Layers in Aligned Large Language Models: The Key to LLM Security.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Safety Layers in Aligned Large Language Models: The Key to LLM Security

Reference 33

Resolution: metadata mismatch; arxiv_id, observed 2026-05-10T12:20:23.070753Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation c21951ba-222c-4ee3-8d29-9d89bd632016 · outbound

This paper cites Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

Reference 34

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:36.984404Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 288a5fb8-4d8c-41d7-add8-f2f1fab2587d · outbound

This paper cites arXiv preprint arXiv:2510.06594 , year=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations arXiv preprint arXiv:2510.06594 , year=

Reference 35

Resolution: verified exact; arxiv_id, observed 2026-05-10T12:20:23.079781Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 1935def0-d11b-4f5c-8a33-228208403b18 · outbound

This paper cites Linearity of Relation Decoding in Transformer Language Models.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Linearity of Relation Decoding in Transformer Language Models

Reference 36

Resolution: metadata mismatch; arxiv_id, observed 2026-05-10T12:20:23.037529Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation da6ffd10-3a93-4014-835a-f3291df9f26e · outbound

This paper cites Eliciting Latent Predictions from Transformers with the Tuned Lens.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Eliciting Latent Predictions from Transformers with the Tuned Lens

Reference 37

Resolution: metadata mismatch; arxiv_id, observed 2026-05-12T16:54:37.831311Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 1a266207-ff1e-4b3c-a046-ba8b81cf6fb0 · outbound

This paper cites Companion Proceedings of the ACM on Web Conference 2025 , pages=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Companion Proceedings of the ACM on Web Conference 2025 , pages=

Reference 38

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:37.005418Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 9a5eb6c6-89f2-4c7a-ae93-07f870c29af0 · outbound

This paper cites Findings of the Association for Computational Linguistics: ACL 2025 , pages=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Findings of the Association for Computational Linguistics: ACL 2025 , pages=

Reference 39

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:36.941560Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 95ba164b-e9b8-4273-83dc-726ed179b39c · outbound

This paper cites The Linear Representation Hypothesis and the Geometry of Large Language Models.

LLM Safety From Within: Detecting Harmful Content with Internal Representations The Linear Representation Hypothesis and the Geometry of Large Language Models

Reference 40

Resolution: metadata mismatch; arxiv_id, observed 2026-05-11T21:43:32.339354Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation fa7e2d14-1c43-4d50-8740-2df308cfcdac · outbound

This paper cites Understanding intermediate layers using linear classifier probes , url =.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Understanding intermediate layers using linear classifier probes , url =

Reference 41

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:36.958245Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation af32c80a-6a32-45b6-a975-6977ccb54369 · outbound

This paper cites An introduction to variable and feature selection , url =.

LLM Safety From Within: Detecting Harmful Content with Internal Representations An introduction to variable and feature selection , url =

Reference 42

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:36.938925Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation ec5f792e-2537-4fb4-a19e-bd90e7521df7 · outbound

This paper cites Qwen3 Technical Report.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Qwen3 Technical Report

Reference 43

Resolution: metadata mismatch; local_arxiv, observed 2026-05-10T12:20:23.019486Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation aa71f58b-386e-47f6-9ee3-b2b19877f419 · outbound

This paper cites arXiv e-prints , pages=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations arXiv e-prints , pages=

Reference 44

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:37.013438Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 9b434071-60ab-45dd-94a1-f0891d6c830a · outbound

This paper cites Lightweight Safety Classification Using Pruned Language Models.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Lightweight Safety Classification Using Pruned Language Models

Reference 45

Resolution: metadata mismatch; arxiv_id, observed 2026-05-10T12:20:23.025851Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 37f1d075-d1a1-44e1-bf1c-d313c220631a · outbound

This paper cites ShieldGemma: Generative AI Content Moderation Based on Gemma.

LLM Safety From Within: Detecting Harmful Content with Internal Representations ShieldGemma: Generative AI Content Moderation Based on Gemma

Reference 46

Resolution: metadata mismatch; arxiv_id, observed 2026-05-20T13:17:39.551920Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation ed317d6a-02e1-4c85-9ecf-89b825591580 · outbound

This paper cites Generative or Discriminative? Revisiting Text Classification in the Era of Transformers.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Generative or Discriminative? Revisiting Text Classification in the Era of Transformers

Reference 47

Resolution: verified exact; doi, observed 2026-05-10T04:35:16.476296Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 4dd6b657-03f0-4045-9038-4df151e4383b · outbound

This paper cites The Thirteenth International Conference on Learning Representations , year=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations The Thirteenth International Conference on Learning Representations , year=

Reference 48

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:37.023071Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 9e0f907f-bdd2-42f2-8de5-68a3b7994bf9 · outbound

This paper cites Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=.

Reference 49

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:36.948506Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 134b24df-8dfe-496c-a19f-abc97a7ff2f8 · outbound

This paper cites IEEE Robotics and Automation Letters , volume=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations IEEE Robotics and Automation Letters , volume=

Reference 50

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:36.978073Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation dff3646d-ff43-490f-af8f-75e286bb77b7 · outbound

This paper cites RDI: An adversarial robustness evaluation metric for deep neural networks based on model statistical features.

LLM Safety From Within: Detecting Harmful Content with Internal Representations RDI: An adversarial robustness evaluation metric for deep neural networks based on model statistical features

Reference 51

Resolution: verified exact; arxiv_id, observed 2026-05-10T12:20:23.060041Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 0352f629-72d9-4450-8d61-fe393b41a693 · outbound

This paper cites Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) , pages=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) , pages=

Reference 52

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:36.944899Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 25396c11-0073-4387-9657-1dcad9e06309 · outbound

This paper cites Scaling laws for neural language models , url =.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Scaling laws for neural language models , url =

Reference 53

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:37.031134Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation ecedd5a7-16f4-4a84-b195-84538c5d991d · outbound

This paper cites Understanding the Effects of RLHF on LLM Generalisation and Diversity.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Understanding the Effects of RLHF on LLM Generalisation and Diversity

Reference 54

Resolution: metadata mismatch; arxiv_id, observed 2026-05-19T02:34:44.407176Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 4b56e116-1bfb-49f2-85bf-d149403d1cdf · outbound

This paper cites author=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations author=

Reference 55

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:37.027719Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation f437287a-2719-4f3a-9179-e74fe41ecda4 · outbound

This paper cites Universal and Transferable Adversarial Attacks on Aligned Language Models.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Universal and Transferable Adversarial Attacks on Aligned Language Models

Reference 56

Resolution: metadata mismatch; local_arxiv, observed 2026-05-10T12:20:23.010454Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 0c8b0a62-c9a2-44f9-b6b3-d4a9fbeb5989 · outbound

This paper cites 2025 , howpublished =.

LLM Safety From Within: Detecting Harmful Content with Internal Representations 2025 , howpublished =

Reference 57

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:36:04.302967Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 9e552a06-6d22-4761-a2d0-2b53260aa24c · outbound

This paper cites 2025 , howpublished =.

LLM Safety From Within: Detecting Harmful Content with Internal Representations 2025 , howpublished =

Reference 58

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:36.951802Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 7b9dbb1e-d773-4a52-ac22-4b31eac9f159 · outbound

This paper cites an unresolved cited work.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Unresolved cited work

Reference 59

Resolution: parse uncertain; raw_fallback, observed 2026-05-22T04:34:36.955330Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 4ffef57a-39c4-491a-8344-0b79e4db3752 · outbound

This paper cites Pooling And Attention: What Are Effective Designs For LLM-Based Embedding Models?.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Pooling And Attention: What Are Effective Designs For LLM-Based Embedding Models?

Reference 60

Resolution: metadata mismatch; arxiv_id, observed 2026-05-10T12:20:23.048764Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 7426ea55-99ba-44c8-a9e5-397ca6e2f056 · outbound

This paper cites DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning.

LLM Safety From Within: Detecting Harmful Content with Internal Representations DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

Reference 61

Resolution: metadata mismatch; local_arxiv, observed 2026-05-10T12:20:23.040333Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 96928c83-816f-4f73-adf8-bb0302d28415 · outbound

This paper cites ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools.

LLM Safety From Within: Detecting Harmful Content with Internal Representations ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools

Reference 62

Resolution: metadata mismatch; arxiv_id, observed 2026-05-11T08:08:10.048088Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 190f9fcb-274b-4fad-91b2-00244340b913 · outbound

This paper cites arXiv preprint arXiv:2508.03550 , year=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations arXiv preprint arXiv:2508.03550 , year=

Reference 63

Resolution: verified exact; arxiv_id, observed 2026-05-10T12:20:23.045949Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation c401dee2-40a4-4c16-a905-26cd98ef813d · outbound

This paper cites Proceedings of the IEEE conference on computer vision and pattern recognition , pages=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

Reference 64

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:36:04.294906Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation fd9f2d22-03f3-4eac-a64c-2d087444a635 · outbound

This paper cites BERT : Pre-training of Deep Bidirectional Transformers for Language Understanding.

LLM Safety From Within: Detecting Harmful Content with Internal Representations BERT : Pre-training of Deep Bidirectional Transformers for Language Understanding

Reference 65

Resolution: verified exact; doi, observed 2026-05-10T04:35:16.486832Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 5b760b1f-10fb-4087-a3cf-f513f2d6604f · outbound

This paper cites 2019 , eprint=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations 2019 , eprint=

Reference 66

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:36:04.298967Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation d0c1504c-2ded-4914-9a9f-f6a8b85423c4 · outbound

This paper cites Hate speech detection and racial bias mitigation in social media based on bert model.PLOS ONE, 15(8):1–26, 08 2020.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Hate speech detection and racial bias mitigation in social media based on bert model.PLOS ONE, 15(8):1–26, 08 2020

Reference 67

Resolution: metadata mismatch; doi, observed 2026-05-10T04:35:16.482488Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation cc4b602a-2457-4275-b6ea-74d178d22fdf · outbound

This paper cites 2021 , isbn =.

LLM Safety From Within: Detecting Harmful Content with Internal Representations 2021 , isbn =

Reference 68

Resolution: verified exact; arxiv_id, observed 2026-05-10T04:35:16.480828Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 10d91182-2d94-4175-8282-2944ecafb995 · outbound

This paper cites H ate BERT : Retraining BERT for Abusive Language Detection in E nglish.

LLM Safety From Within: Detecting Harmful Content with Internal Representations H ate BERT : Retraining BERT for Abusive Language Detection in E nglish

Reference 69

Resolution: verified exact; doi, observed 2026-05-10T04:35:16.478300Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 02b07d0d-af88-48fd-8079-717a37ffa34a · outbound

This paper cites 2024 , eprint=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations 2024 , eprint=

Reference 70

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:36:04.306299Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 06a6ec07-824a-4b46-8aa8-5c30d572501c · outbound

This paper cites and Tay, Yi and Sorensen, Jeffrey and Gupta, Jai and Metzler, Donald and Vasserman, Lucy , title =.

LLM Safety From Within: Detecting Harmful Content with Internal Representations and Tay, Yi and Sorensen, Jeffrey and Gupta, Jai and Metzler, Donald and Vasserman, Lucy , title =

Reference 71

Resolution: metadata mismatch; arxiv_id, observed 2026-05-10T04:35:16.485012Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 5abbd1ea-32e1-4861-b346-9b5e7c58d9c5 · outbound

This paper cites 2025 , eprint=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations 2025 , eprint=

Reference 72

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:34:37.034244Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 0421a101-051c-442a-a6b4-4b3a122c85d1 · outbound

This paper cites The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

Reference 73

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:36:04.309711Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 1e9acd6d-91b0-4870-a830-94de92d2b6cd · outbound

This paper cites PolyGuard: A Multilingual Safety Moderation Tool for 17 Languages.

LLM Safety From Within: Detecting Harmful Content with Internal Representations PolyGuard: A Multilingual Safety Moderation Tool for 17 Languages

Reference 74

Resolution: metadata mismatch; arxiv_id, observed 2026-05-10T12:20:23.043286Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 1a853100-0621-4b91-8431-916beeb54807 · outbound

This paper cites Linear Representations of Sentiment in Large Language Models.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Linear Representations of Sentiment in Large Language Models

Reference 75

Resolution: metadata mismatch; arxiv_id, observed 2026-05-15T12:43:06.705954Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 36b3f119-44c3-4de0-8410-ade08be64a1a · outbound

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

LLM Safety From Within: Detecting Harmful Content with Internal Representations The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets

Reference 76

Resolution: metadata mismatch; arxiv_id, observed 2026-05-12T19:35:39.400821Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 279548d1-4dae-4d91-9093-32759b537cbf · outbound

This paper cites Proceedings of the 28th ACM international conference on information and knowledge management , pages=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Proceedings of the 28th ACM international conference on information and knowledge management , pages=

Reference 77

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:36:04.315979Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation f2aa0f24-f7bf-4c7f-9eb9-2c042ce9d75b · outbound

This paper cites arXiv preprint arXiv:2510.18081 , year=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations arXiv preprint arXiv:2510.18081 , year=

Reference 78

Resolution: verified exact; arxiv_id, observed 2026-05-10T12:20:23.065573Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 9b0a7544-a00d-4285-9474-24c935d534b5 · outbound

This paper cites Curvalid: Geometrically-guided adversarial prompt detection.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Curvalid: Geometrically-guided adversarial prompt detection

Reference 79

Resolution: metadata mismatch; arxiv_id, observed 2026-05-10T12:20:23.073944Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 79fa70d3-2184-45cb-b67e-8ed50c7fd44f · outbound

This paper cites Neural networks , volume=.

LLM Safety From Within: Detecting Harmful Content with Internal Representations Neural networks , volume=

Reference 80

Resolution: verified fuzzy; raw_fallback, observed 2026-05-22T04:36:04.312630Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Pith citing papers

Observation 85e0dbd5-f316-43eb-a156-6af8652f9f7f · inbound

MINER: Mining Multimodal Internal Representation for Efficient Retrieval cites this paper.

MINER: Mining Multimodal Internal Representation for Efficient Retrieval LLM Safety From Within: Detecting Harmful Content with Internal Representations

Reference 16

Resolution: verified exact; local_arxiv, observed 2026-05-11T19:06:10.607800Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 89bd88d0-aa8e-4cca-80eb-23bae07202bb · inbound

LCGuard: Latent Communication Guard for Safe KV Sharing in Multi-Agent Systems cites this paper.

LCGuard: Latent Communication Guard for Safe KV Sharing in Multi-Agent Systems LLM Safety From Within: Detecting Harmful Content with Internal Representations

Reference 21

Resolution: verified exact; local_arxiv, observed 2026-05-22T05:01:05.907482Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 07051512-4efa-4e18-b1c7-73f64e217c15 · inbound

AERIC: Anticipatory Hidden-State Monitoring for Implicit Harmful Dialogue cites this paper.

AERIC: Anticipatory Hidden-State Monitoring for Implicit Harmful Dialogue LLM Safety From Within: Detecting Harmful Content with Internal Representations

Reference 15

Resolution: metadata mismatch; local_arxiv, observed 2026-06-30T21:35:04.762580Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.