Paper Citation Record · LEDGER

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense

As of 18 August 2026, this Paper Citation Record lists 77 of 77 outbound references and 6 inbound Pith citation observations for arXiv:2411.08410.

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

pith.paper-citation-record.v1
2411.08410 v2

Coverage vector

measured 77 of 77 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-08-12T21:45:34.938217Z

measured 83 of 83 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-17T06:30:58.91139+00:00

measured 6 of 6 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links, observed 2026-08-07T19:45:19.069073Z

measured 0 of 1 external citation measurements

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

Source: arxiv_reference, observed 2026-05-23T20:58:26.217531Z

Reference resolution

77 of 77 outbound references displayed

verified exact0
verified fuzzy36
unresolved41
parse uncertain0
malformed identifier0
metadata mismatch0

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation ecacfeb8-4c19-4d9e-9be2-96fc98f7f78a · outbound

This paper cites Gpt-4v(ision) system card.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Gpt-4v(ision) system card

Reference 1

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:36.092236Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 06b042ef-f91c-4bad-8638-f6d3f1e659b3 · outbound

This paper cites Hello gpt-4o.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Hello gpt-4o

Reference 2

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:36.075844Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation f21b1d87-a7b1-4e72-8eb8-3c11cc7c3def · outbound

This paper cites Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback

Reference 3

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.611296Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3461a50c-81c5-4dda-91ea-541307814ad3 · outbound

This paper cites Red-Teaming Large Language Models using Chain of Utterances for Safety-Alignment.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Red-Teaming Large Language Models using Chain of Utterances for Safety-Alignment

Reference 4

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.616818Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3493064a-40a5-4543-af27-b85a26c9730b · outbound

This paper cites Safety-tuned llamas: Lessons from improving the safety of large language models that follow instructions.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Safety-tuned llamas: Lessons from improving the safety of large language models that follow instructions

Reference 5

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:36.061703Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 54ecac06-d7c2-41c8-8be1-aeeea2889bf4 · outbound

This paper cites Choquette- Choo, Matthew Jagielski, Irena Gao, Pang Wei Koh, Daphne Ippolito, Florian Tram`er, and Ludwig Schmidt.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Choquette- Choo, Matthew Jagielski, Irena Gao, Pang Wei Koh, Daphne Ippolito, Florian Tram`er, and Ludwig Schmidt

Reference 6

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:36.046567Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 2361ca49-7a80-4342-8248-e8c08959c726 · outbound

This paper cites Pap- pas, Florian Tram `er, Hamed Hassani, and Eric Wong.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Pap- pas, Florian Tram `er, Hamed Hassani, and Eric Wong

Reference 7

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:36.032512Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation c1d2fff6-07af-4609-8bf5-b4ee836295f2 · outbound

This paper cites MiniGPT-v2: large language model as a unified interface for vision-language multi-task learning.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense MiniGPT-v2: large language model as a unified interface for vision-language multi-task learning

Reference 8

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.634774Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation a20fc002-86f6-4e91-adbf-0801375658d8 · outbound

This paper cites Can Language Models be Instructed to Protect Personal Information?.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Can Language Models be Instructed to Protect Personal Information?

Reference 9

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.638955Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 06afd5cf-7928-4a1a-8714-6525907d8fbc · outbound

This paper cites DRESS : Instructing large vision-language models to align and interact with humans via natural lan- guage feedback.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense DRESS : Instructing large vision-language models to align and interact with humans via natural lan- guage feedback

Reference 10

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:36.017777Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation fdb1bf01-0868-41aa-b276-6fdaf1a9ddf8 · outbound

This paper cites InternVL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Tasks.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense InternVL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Tasks

Reference 11

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.647230Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9bf26ada-9e7e-4445-a634-ac1ac8a9ddf3 · outbound

This paper cites Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference

Reference 12

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.652008Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c4d1032d-7ef6-44eb-9255-e79752111b20 · outbound

This paper cites A coefficient of agreement for nominal scales.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense A coefficient of agreement for nominal scales

Reference 13

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.656566Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 572e37af-34a5-4376-8088-8056e0d120ce · outbound

This paper cites Attack prompt generation for red teaming and defending large language models.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Attack prompt generation for red teaming and defending large language models

Reference 14

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.993565Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 09e90073-9bf4-489f-9db3-5518c7f206cc · outbound

This paper cites Multilingual jailbreak challenges in large language models.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Multilingual jailbreak challenges in large language models

Reference 15

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.980111Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 7a1fe5cc-214a-4d81-832a-80db6e1947c4 · outbound

This paper cites The Llama 3 Herd of Models.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense The Llama 3 Herd of Models

Reference 16

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.668351Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ba059300-2e6e-44dc-801d-8f4ed4ffd5c4 · outbound

This paper cites Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned

Reference 17

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.672975Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d7e3f681-6b8a-449c-a3be-074ed6c5480a · outbound

This paper cites Inducing high energy-latency of large vision-language models with verbose images.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Inducing high energy-latency of large vision-language models with verbose images

Reference 18

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.965845Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation f590a976-a726-4e93-a67b-bcf63b63f027 · outbound

This paper cites FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts

Reference 19

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.681816Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation a68c2505-6927-46f8-b130-1ac02d4c3e0b · outbound

This paper cites Eyes Closed, Safety On: Protecting Multimodal LLMs via Image-to-Text Transformation.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Eyes Closed, Safety On: Protecting Multimodal LLMs via Image-to-Text Transformation

Reference 20

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.686526Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3fca6842-0e57-40bd-934d-98fe1b56ec11 · outbound

This paper cites Agent smith: A single image can jailbreak one million multimodal LLM agents exponentially fast.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Agent smith: A single image can jailbreak one million multimodal LLM agents exponentially fast

Reference 21

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.950361Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation cc83ef47-471d-4002-9e77-0837e729f817 · outbound

This paper cites UNK-VQA: A Dataset and a Probe into the Abstention Ability of Multi-modal Large Models.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense UNK-VQA: A Dataset and a Probe into the Abstention Ability of Multi-modal Large Models

Reference 22

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.696102Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 17098e69-3fa0-493a-bf05-ac2db890ea93 · outbound

This paper cites Catastrophic jailbreak of open-source llms via exploiting generation.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Catastrophic jailbreak of open-source llms via exploiting generation

Reference 23

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.934468Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 8324f36b-7883-4813-a441-dca17ffcc30d · outbound

This paper cites an unresolved cited work.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Unresolved cited work

Reference 24

Resolution: unresolved; raw_fallback, observed 2026-08-12T21:45:35.917928Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 707a7516-2481-49a6-afa5-92b3df368c85 · outbound

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

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations

Reference 25

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.709820Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 7d3799cc-10a3-4da0-a305-fe5063fa8249 · outbound

This paper cites Mistral 7B.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Mistral 7B

Reference 26

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.715063Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 1207463b-f4c6-43e8-9168-9e3c94f3d2fe · outbound

This paper cites Dragan, Aditi Raghunathan, and Jacob Steinhardt.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Dragan, Aditi Raghunathan, and Jacob Steinhardt

Reference 27

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.902586Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation b4b0bba9-5125-4bb9-8fd2-0fe18e85ba24 · outbound

This paper cites Challenges and Applications of Large Language Models.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Challenges and Applications of Large Language Models

Reference 28

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.724487Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d8447208-f019-47f7-82a9-87ec358b5985 · outbound

This paper cites LoRA Fine-tuning Efficiently Undoes Safety Training in Llama 2-Chat 70B.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense LoRA Fine-tuning Efficiently Undoes Safety Training in Llama 2-Chat 70B

Reference 29

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.729143Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation eeeea511-4dec-4adb-9a37-85629996b12d · outbound

This paper cites Red teaming visual language models.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Red teaming visual language models

Reference 30

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.887647Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 3d9e65fa-d1bf-486f-b6f2-106ce51141c8 · outbound

This paper cites Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll ´ar, and C.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll ´ar, and C

Reference 31

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.872910Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 8be0ef91-97df-4f21-bfb1-19d118d458bc · outbound

This paper cites A Survey of Attacks on Large Vision-Language Models: Resources, Advances, and Future Trends.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense A Survey of Attacks on Large Vision-Language Models: Resources, Advances, and Future Trends

Reference 32

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.742195Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 323f1df9-e28e-4e38-95a3-ee8f874db4e0 · outbound

This paper cites Improved baselines with visual instruction tuning.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Improved baselines with visual instruction tuning

Reference 33

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.858655Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation ebe34c91-1813-4f2a-97ea-3f307c438de6 · outbound

This paper cites Llava-next: Im- proved reasoning, ocr, and world knowledge, 2024.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Llava-next: Im- proved reasoning, ocr, and world knowledge, 2024

Reference 34

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.750201Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 8c5e5d15-26a8-41bf-ab98-b685393a9d97 · outbound

This paper cites Jailbreak Attacks and Defenses against Multimodal Generative Models: A Survey.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Jailbreak Attacks and Defenses against Multimodal Generative Models: A Survey

Reference 35

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.754730Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 4643bf07-a145-4787-a14e-072d401c007c · outbound

This paper cites Mm-safetybench: A benchmark for safety eval- uation of multimodal large language models, 2024.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Mm-safetybench: A benchmark for safety eval- uation of multimodal large language models, 2024

Reference 36

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.835192Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 02be041f-a0cf-41e5-9b92-eaaae0c59d96 · outbound

This paper cites AutoJailbreak: Exploring Jailbreak Attacks and Defenses through a Dependency Lens.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense AutoJailbreak: Exploring Jailbreak Attacks and Defenses through a Dependency Lens

Reference 37

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.762910Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 617a036d-0f17-4b5c-957c-28b96eb7e7e1 · outbound

This paper cites An image is worth 1000 lies: Transferability of adversarial images across prompts on vision-language models.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense An image is worth 1000 lies: Transferability of adversarial images across prompts on vision-language models

Reference 38

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.820591Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 40cc0e55-5bd5-4c7b-b544-9fb36324f386 · outbound

This paper cites JailBreakV: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attacks.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense JailBreakV: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attacks

Reference 39

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.771126Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 898fa128-a27d-418d-a5f8-983e54cc1ff1 · outbound

This paper cites Towards deep learning 10 models resistant to adversarial attacks.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Towards deep learning 10 models resistant to adversarial attacks

Reference 40

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.806611Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 1218dd34-37fa-46b6-8471-bf137b965037 · outbound

This paper cites Rule based rewards for fine-grained LLM safety.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Rule based rewards for fine-grained LLM safety

Reference 41

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.791374Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 1a57df71-d351-4954-b50d-fd5516a3299c · outbound

This paper cites GPT-4 Technical Report.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense GPT-4 Technical Report

Reference 42

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.783693Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 77a18ca8-b8a5-4a26-ac1e-706b6816276a · outbound

This paper cites an unresolved cited work.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Unresolved cited work

Reference 43

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.788025Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation dec6bacb-b5c1-4f9e-aac0-177d634eb73d · outbound

This paper cites LLM improvement for jailbreak defense: Analysis through the lens of over-refusal.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense LLM improvement for jailbreak defense: Analysis through the lens of over-refusal

Reference 44

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.765765Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation fe8bb108-1dcf-4a4d-858e-0bde1428ddff · outbound

This paper cites Francis Song, Trevor Cai, Roman Ring, John Aslanides, Amelia Glaese, Nat McAleese, and Geoffrey Irving.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Francis Song, Trevor Cai, Roman Ring, John Aslanides, Amelia Glaese, Nat McAleese, and Geoffrey Irving

Reference 45

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.748812Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 1e1abc64-67fc-4d23-8dad-1e17b76af5d3 · outbound

This paper cites MLLM-Protector: Ensuring MLLM's Safety without Hurting Performance.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense MLLM-Protector: Ensuring MLLM's Safety without Hurting Performance

Reference 46

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.801256Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation dd3ccc18-da4a-47fb-9f13-017a5cc90c4c · outbound

This paper cites Visual adversarial examples jailbreak aligned large language models.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Visual adversarial examples jailbreak aligned large language models

Reference 47

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.733234Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 938ffd8f-06ee-4800-8778-96972370ddb3 · outbound

This paper cites Fine-tuning aligned language models compromises safety, even when users do not intend to! In ICLR.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Fine-tuning aligned language models compromises safety, even when users do not intend to! In ICLR

Reference 48

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.717801Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 66775bff-8661-4800-b36e-b08b4ff27f80 · outbound

This paper cites High-resolution image syn- thesis with latent diffusion models.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense High-resolution image syn- thesis with latent diffusion models

Reference 49

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.701375Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation cce7922a-31ea-4c0b-b708-04ccb18b81e7 · outbound

This paper cites On the adversar- ial robustness of multi-modal foundation models.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense On the adversar- ial robustness of multi-modal foundation models

Reference 50

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.686400Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation b76417da-dda9-4d8a-b81a-e7c48dff8776 · outbound

This paper cites Ignore This Title and HackAPrompt: Exposing Systemic Vulnerabilities of LLMs through a Global Scale Prompt Hacking Competition.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Ignore This Title and HackAPrompt: Exposing Systemic Vulnerabilities of LLMs through a Global Scale Prompt Hacking Competition

Reference 51

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.823771Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 308a26f8-127f-4afb-bf41-37085e964dd7 · outbound

This paper cites SPML: A DSL for Defending Language Models Against Prompt Attacks.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense SPML: A DSL for Defending Language Models Against Prompt Attacks

Reference 52

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.828353Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ff7bc135-3189-43b4-9a6a-ac419c2beac1 · outbound

This paper cites Abu-Ghazaleh.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Abu-Ghazaleh

Reference 53

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.833016Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 6b7d7880-c659-439c-a442-6e1721fe630e · outbound

This paper cites Hugginggpt: Solving AI tasks with chatgpt and its friends in hugging face.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Hugginggpt: Solving AI tasks with chatgpt and its friends in hugging face

Reference 54

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.661999Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 408bbfb6-464b-4448-86be-c5866abc4dd6 · outbound

This paper cites Assessment of Multimodal Large Language Models in Alignment with Human Values.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Assessment of Multimodal Large Language Models in Alignment with Human Values

Reference 55

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.842506Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 967bc02b-2930-47f5-8d99-96b4c6eae9c2 · outbound

This paper cites Qwen2.5: A party of foundation models, 2024.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Qwen2.5: A party of foundation models, 2024

Reference 56

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.647386Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 1b73fded-eb8a-4d2a-b947-5244254832fa · outbound

This paper cites Llama 2: Open Foundation and Fine-Tuned Chat Models.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Llama 2: Open Foundation and Fine-Tuned Chat Models

Reference 57

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.851653Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 1b95a681-39a3-4b64-b76f-c96ad0bfd364 · outbound

This paper cites The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions

Reference 58

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.856687Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 48e842a5-6044-42b7-ac68-45373c43951c · outbound

This paper cites Truong, Simran Arora, Mantas Mazeika, Dan Hendrycks, Zinan Lin, Yu Cheng, Sanmi Koyejo, Dawn Song, and Bo Li.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Truong, Simran Arora, Mantas Mazeika, Dan Hendrycks, Zinan Lin, Yu Cheng, Sanmi Koyejo, Dawn Song, and Bo Li

Reference 59

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.861227Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 6afee372-28f2-46b0-bb00-646abb2ba6fb · outbound

This paper cites Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution, 2024.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution, 2024

Reference 60

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.623184Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 1d3c8071-37b8-4863-8081-72cde92c513e · outbound

This paper cites InferAligner: Inference-Time Alignment for Harmlessness through Cross-Model Guidance.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense InferAligner: Inference-Time Alignment for Harmlessness through Cross-Model Guidance

Reference 61

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.868801Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9675dc97-a4d3-4eea-8e3f-3411d8ca5b70 · outbound

This paper cites White-box multimodal jailbreaks against large vision-language models.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense White-box multimodal jailbreaks against large vision-language models

Reference 62

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.609900Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation d6ae2c1f-32de-4681-ad24-9ada369b9ba3 · outbound

This paper cites CogVLM: Visual Expert for Pretrained Language Models.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense CogVLM: Visual Expert for Pretrained Language Models

Reference 63

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.876990Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation adcd44df-1b37-43a0-a11f-a22db48f68c6 · outbound

This paper cites AdaShield: Safeguarding Multimodal Large Language Models from Structure-based Attack via Adaptive Shield Prompting.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense AdaShield: Safeguarding Multimodal Large Language Models from Structure-based Attack via Adaptive Shield Prompting

Reference 64

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.881362Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation b02b3b48-86d9-48c0-9aef-86fdac91ee19 · outbound

This paper cites Jailbreaking GPT-4V via Self-Adversarial Attacks with System Prompts.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Jailbreaking GPT-4V via Self-Adversarial Attacks with System Prompts

Reference 65

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.885499Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 51870101-dccd-4efd-a82c-5571ce587b8f · outbound

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

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Jailbreak Attacks and Defenses Against Large Language Models: A Survey

Reference 66

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.890372Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 4c697ac4-a4c4-4339-b60a-612a049eb5c8 · outbound

This paper cites GPTFUZZER: Red Teaming Large Language Models with Auto-Generated Jailbreak Prompts.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense GPTFUZZER: Red Teaming Large Language Models with Auto-Generated Jailbreak Prompts

Reference 67

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.894689Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 7a342db7-75cb-4652-a1cf-0f94bc605333 · outbound

This paper cites Mm-vet: Evaluating large multimodal models for integrated capabilities.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Mm-vet: Evaluating large multimodal models for integrated capabilities

Reference 68

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.595763Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation bb8fb085-a648-4b80-8a97-22fcbf5f518e · outbound

This paper cites GPT- 4 is too smart to be safe: Stealthy chat with llms via cipher.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense GPT- 4 is too smart to be safe: Stealthy chat with llms via cipher

Reference 69

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.580839Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation cdbbd4c5-9245-4072-8461-e60da76bb17a · outbound

This paper cites Removing RLHF pro- tections in GPT-4 via fine-tuning.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Removing RLHF pro- tections in GPT-4 via fine-tuning

Reference 70

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.564389Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 9ace8bd1-6ad4-4c42-89aa-d556e2686950 · outbound

This paper cites Make Them Spill the Beans! Coercive Knowledge Extraction from (Production) LLMs.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Make Them Spill the Beans! Coercive Knowledge Extraction from (Production) LLMs

Reference 71

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.911487Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation be182522-4a0a-4527-a7f3-54bcef14df24 · outbound

This paper cites On evaluating adversarial robustness of large vision-language models.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense On evaluating adversarial robustness of large vision-language models

Reference 72

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.548999Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation a1ff2074-240c-43fd-afc2-41a9813f044a · outbound

This paper cites Xing, Hao Zhang, Joseph E.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Xing, Hao Zhang, Joseph E

Reference 73

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.534002Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 9f54ee16-64f1-4909-ad3d-03e40485bfea · outbound

This paper cites Autodan: Interpretable gradient-based adversarial attacks on large language models.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Autodan: Interpretable gradient-based adversarial attacks on large language models

Reference 74

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.518731Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 6acc45e3-c44f-4f90-a700-3068ea37b819 · outbound

This paper cites Hospedales.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Hospedales

Reference 75

Resolution: verified fuzzy; raw_fallback, observed 2026-08-12T21:45:35.502054Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 41bceb35-d66f-43dc-9492-9fca20954944 · outbound

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

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Universal and Transferable Adversarial Attacks on Aligned Language Models

Reference 76

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.933478Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9e01f9fc-a7e2-4386-b021-52139ee90f8e · outbound

This paper cites Is the System Message Really Important to Jailbreaks in Large Language Models?.

The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense Is the System Message Really Important to Jailbreaks in Large Language Models?

Reference 77

Resolution: unresolved; no resolver link, observed 2026-08-12T21:45:34.938217Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Pith citing papers

Observation fadbe50f-fa67-4b90-980f-4779dfbe4c43 · inbound

Harmful Fine-tuning Attacks and Defenses for Large Language Models: A Survey cites this paper.

Harmful Fine-tuning Attacks and Defenses for Large Language Models: A Survey The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense

Reference 47

Resolution: verified exact; arxiv_id, observed 2026-05-23T20:58:26.220359Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation b203b1e8-2def-4037-b255-75603495a01a · inbound

A Survey of Safety on Large Vision-Language Models: Attacks, Defenses and Evaluations cites this paper.

A Survey of Safety on Large Vision-Language Models: Attacks, Defenses and Evaluations The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense

Reference 72

Resolution: unresolved; no resolver link, observed 2026-08-07T19:45:19.069073Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 6238b638-3e4c-4bf3-beb4-1af93b5b1163 · inbound

Seeing the Threat: Vulnerabilities in Vision-Language Models to Adversarial Attack cites this paper.

Seeing the Threat: Vulnerabilities in Vision-Language Models to Adversarial Attack The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense

Reference 11

Resolution: unresolved; no resolver link, observed 2026-08-07T13:22:01.083283Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation e9266238-cdb5-4028-8db9-a8b02f5be4f2 · inbound

Mitigating Object Hallucination via Robust Local Perception Search cites this paper.

Mitigating Object Hallucination via Robust Local Perception Search The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense

Reference 14

Resolution: unresolved; no resolver link, observed 2026-08-07T05:53:59.867495Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9638a436-a417-4c85-8988-bdbaa52ec948 · inbound

Secure Tug-of-War (SecTOW): Iterative Defense-Attack Training with Reinforcement Learning for Multimodal Model Security cites this paper.

Secure Tug-of-War (SecTOW): Iterative Defense-Attack Training with Reinforcement Learning for Multimodal Model Security The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense

Reference 17

Resolution: unresolved; no resolver link, observed 2026-08-06T12:09:39.684858Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9adf6efa-39f9-4f04-a303-024b5adda4f7 · inbound

VERA-V: Variational Inference Framework for Jailbreaking Vision-Language Models cites this paper.

VERA-V: Variational Inference Framework for Jailbreaking Vision-Language Models The VLLM Safety Paradox: Dual Ease in Jailbreak Attack and Defense

Reference 8

Resolution: unresolved; no resolver link, observed 2026-08-04T09:02:37.531373Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.