Paper Citation Record · LEDGER

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space

As of 9 August 2026, this Paper Citation Record lists 33 of 33 outbound references and 0 inbound Pith citation observations for arXiv:2607.15218.

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

pith.paper-citation-record.v1
2607.15218 v1

Coverage vector

measured 33 of 33 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-08-01T23:52:53.822416Z

measured 33 of 33 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-08T06:32:00.761636+00:00

measured 0 of 0 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links

measured 0 of 1 external citation measurements

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

Source: cited_works

Reference resolution

33 of 33 outbound references displayed

verified exact0
verified fuzzy0
unresolved33
parse uncertain0
malformed identifier0
metadata mismatch0

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation 1fc52820-4435-49d1-be5c-a05ae20aca2e · outbound

This paper cites Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone

Reference 1

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:50.942964Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 99c3a7f3-b077-4896-a13a-3bcc5206d674 · outbound

This paper cites Constitutional AI: Harmlessness from AI Feedback.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space Constitutional AI: Harmlessness from AI Feedback

Reference 4

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:51.152417Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 8e25b957-f304-4cbb-9607-cd4b12107e11 · outbound

This paper cites SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model

Reference 5

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:51.225771Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3c92058b-21ed-4451-9603-3beaa4b563dd · outbound

This paper cites an unresolved cited work.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space Unresolved cited work

Reference 9

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:51.548149Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 0a908996-fc3e-49fb-9ad0-651b6a13bb62 · outbound

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

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space AEGIS: Online Adaptive AI Content Safety Moderation with Ensemble of LLM Experts

Reference 10

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:51.607980Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3e103e25-3104-4283-b76d-e040d337b233 · outbound

This paper cites A Framework for Benchmarking and Aligning Task-Planning Safety in LLM-Based Embodied Agents.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space A Framework for Benchmarking and Aligning Task-Planning Safety in LLM-Based Embodied Agents

Reference 11

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:51.679091Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation b9897fd6-cab1-43d9-af9a-a2a9f498b85d · outbound

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

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations

Reference 12

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:51.813389Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f307715a-0c52-4172-bcb3-d55374f2fb78 · outbound

This paper cites AI2-THOR: An Interactive 3D Environment for Visual AI.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space AI2-THOR: An Interactive 3D Environment for Visual AI

Reference 13

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:51.884733Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 45d32663-42a5-4dc4-af10-23bdb3dd896e · outbound

This paper cites AGENTSAFE: Benchmarking the safety of embodied agents on hazardous instructions.arXiv preprint arXiv:2506.14697,.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space AGENTSAFE: Benchmarking the safety of embodied agents on hazardous instructions.arXiv preprint arXiv:2506.14697,

Reference 15

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:51.980971Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 04633014-3ac9-4779-ad91-b89bf0da1569 · outbound

This paper cites G-Eval: NLG evaluation using GPT-4 with better human alignment.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space G-Eval: NLG evaluation using GPT-4 with better human alignment

Reference 16

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:52.049845Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 99da4540-4375-4875-b54d-c7697ac349a4 · outbound

This paper cites The Llama 3 Herd of Models.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space The Llama 3 Herd of Models

Reference 17

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:52.190795Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation e51a3e38-95ef-4544-b86e-85866116c1f4 · outbound

This paper cites IS-Bench: Evaluating interactive safety of VLM-driven embodied agents in daily household tasks.arXiv preprint arXiv:2506.16402,.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space IS-Bench: Evaluating interactive safety of VLM-driven embodied agents in daily household tasks.arXiv preprint arXiv:2506.16402,

Reference 18

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:52.274400Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3da758d5-4aee-449c-819c-4ebfcceacde1 · outbound

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

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets

Reference 19

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:52.384452Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 5683ea90-6138-47ea-bb6e-be781e9317ad · outbound

This paper cites Don’t let your robot be harmful: Responsible robotic manipulation via safety- as-policy.arXiv preprint arXiv:2411.18289,.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space Don’t let your robot be harmful: Responsible robotic manipulation via safety- as-policy.arXiv preprint arXiv:2411.18289,

Reference 20

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:52.538702Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 22e84c45-dd55-4ab4-baf2-a22da5c9bca9 · outbound

This paper cites SafePlan: Leveraging Formal Logic and Chain-of-Thought Reasoning for Enhanced Safety in LLM-based Robotic Task Planning.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space SafePlan: Leveraging Formal Logic and Chain-of-Thought Reasoning for Enhanced Safety in LLM-based Robotic Task Planning

Reference 21

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:52.704841Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 24fd2da0-bf5b-4d9c-87f0-2bea544254ca · outbound

This paper cites GPT-4 Technical Report.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space GPT-4 Technical Report

Reference 22

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:52.872352Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d98b5c67-da3e-4355-bc3b-be410e503ddb · outbound

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

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space Llama 2: Open Foundation and Fine-Tuned Chat Models

Reference 25

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:53.302816Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 5ec62ffe-87e2-4754-ba2d-9e91bed1400e · outbound

This paper cites Steering Language Models With Activation Engineering.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space Steering Language Models With Activation Engineering

Reference 26

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:53.373741Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 1e1e2cb9-250e-4926-b811-27d758c4e5ab · outbound

This paper cites Voyager: An Open-Ended Embodied Agent with Large Language Models.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space Voyager: An Open-Ended Embodied Agent with Large Language Models

Reference 27

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:53.449275Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation b8684d64-8159-4a6d-a6ac-07c8750107e7 · outbound

This paper cites Advancing Embodied Agent Security: From Safety Benchmarks to Input Moderation.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space Advancing Embodied Agent Security: From Safety Benchmarks to Input Moderation

Reference 28

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:53.514874Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 490fb07f-1faa-4fbf-abf3-7c9a68b1cd51 · outbound

This paper cites SafeAgentBench: A benchmark for safe task planning of embodied LLM agents.arXiv preprint arXiv:2412.13178,.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space SafeAgentBench: A benchmark for safe task planning of embodied LLM agents.arXiv preprint arXiv:2412.13178,

Reference 29

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:53.574727Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation b77de546-c204-434f-b5c5-7b7d79ddacb7 · outbound

This paper cites R-Judge: Bench- marking safety risk awareness for LLM agents.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space R-Judge: Bench- marking safety risk awareness for LLM agents

Reference 30

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:53.629239Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation bf547adc-49f9-4247-a269-2ae631839471 · outbound

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

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space ShieldGemma: Generative AI Content Moderation Based on Gemma

Reference 31

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:53.689569Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 02dd3a31-fca3-4fbf-8a7b-fb2e2aa900c6 · outbound

This paper cites EARBench: Towards Evaluating Physical Risk Awareness for Task Planning of Foundation Model-based Embodied AI Agents.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space EARBench: Towards Evaluating Physical Risk Awareness for Task Planning of Foundation Model-based Embodied AI Agents

Reference 32

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:53.743894Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 1b5897e3-f3dd-407e-934d-6791ca3f3427 · outbound

This paper cites Representation Engineering: A Top-Down Approach to AI Transparency.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space Representation Engineering: A Top-Down Approach to AI Transparency

Reference 33

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:53.822416Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d07c7891-e54a-4e97-803f-9b342e3d8767 · outbound

This paper cites SafeText: A benchmark for exploring physical safety in language models.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space SafeText: A benchmark for exploring physical safety in language models

Reference 2017

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:51.934761Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 45d71624-eabe-449b-aca1-166d013169a5 · outbound

This paper cites Qwen2.5 Technical Report.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space Qwen2.5 Technical Report

Reference 2018

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:53.025835Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 7929156b-5393-4500-9fa6-acfdce7b5cea · outbound

This paper cites Generating Robot Constitutions & Benchmarks for Semantic Safety.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space Generating Robot Constitutions & Benchmarks for Semantic Safety

Reference 2019

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:53.215213Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d6268bef-04c6-4401-a31a-56d243b830a0 · outbound

This paper cites RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control

Reference 2020

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:51.313791Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 1af75e5c-fa88-41ee-8bc1-d41f74c9ba51 · outbound

This paper cites Understanding intermediate layers using linear classifier probes.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space Understanding intermediate layers using linear classifier probes

Reference 2022

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:51.078723Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 6ee6a447-c712-4375-98d0-1a3d8f75ca07 · outbound

This paper cites SafeMind: Benchmarking and mitigating safety risks in embodied LLM agents.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space SafeMind: Benchmarking and mitigating safety risks in embodied LLM agents

Reference 2023

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:51.422778Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 01797a0e-f032-45d7-a546-b4681acb2661 · outbound

This paper cites Do As I Can, Not As I Say: Grounding Language in Robotic Affordances.

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space Do As I Can, Not As I Say: Grounding Language in Robotic Affordances

Reference 2024

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:51.006565Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 520e3158-13b0-453c-9911-a67b79e05fe0 · outbound

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

When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space Efficient LLM Moderation with Multi-Layer Latent Prototypes

Reference 2025

Resolution: unresolved; no resolver link, observed 2026-08-01T23:52:51.468348Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Pith citing papers

No inbound Pith citation observations are available.