Paper Citation Record · LEDGER

From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

As of 14 August 2026, this Paper Citation Record lists 6 of 6 outbound references and 21 inbound Pith citation observations for arXiv:2508.09224.

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

pith.paper-citation-record.v1
2508.09224 v1

Coverage vector

measured 6 of 6 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-08-05T21:35:34.616461Z

measured 27 of 27 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-14T06:32:32.682623+00:00

measured 21 of 21 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links, observed 2026-08-14T04:34:57.214509Z

measured 1 of 1 external citation measurements

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

Source: arxiv_reference, observed 2026-08-05T02:28:24.338817Z

Reference resolution

6 of 6 outbound references displayed

verified exact0
verified fuzzy1
unresolved5
parse uncertain0
malformed identifier0
metadata mismatch0

External citation measurements

0
arxiv_reference, observed 2026-08-05T02:28:24.338817Z

Outbound references

Observation 6fae85d6-77d2-416f-8c51-c429779c4d98 · outbound

This paper cites Jailbreaking Leading Safety-Aligned LLMs with Simple Adaptive Attacks.

From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training Jailbreaking Leading Safety-Aligned LLMs with Simple Adaptive Attacks

Reference 1

Resolution: unresolved; no resolver link, observed 2026-08-05T21:35:33.950466Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f2be2c8c-8ac3-4957-8e84-5c49decdb846 · outbound

This paper cites [14]OpenAI.

From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training [14]OpenAI

Reference 5

Resolution: verified fuzzy; raw_fallback, observed 2026-08-05T21:35:34.923470Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.

Observation 81b07747-e930-4ec7-94d4-ab893594fb53 · outbound

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

From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models

Reference 6

Resolution: unresolved; no resolver link, observed 2026-08-05T21:35:34.494480Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9c3d15c9-ca69-4187-af43-76ddc9f34c93 · outbound

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

From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback

Reference 2022

Resolution: unresolved; no resolver link, observed 2026-08-05T21:35:34.096588Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation e05cc146-80ba-4697-b710-292789aa3a39 · outbound

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

From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training Universal and Transferable Adversarial Attacks on Aligned Language Models

Reference 2023

Resolution: unresolved; no resolver link, observed 2026-08-05T21:35:34.616461Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9a63e92e-2b3c-4b20-afb6-6ea4a7b1b4db · outbound

This paper cites GPT-4o System Card.

From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training GPT-4o System Card

Reference 2024

Resolution: unresolved; no resolver link, observed 2026-08-05T21:35:34.168865Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Pith citing papers

Observation d5e3815c-4641-4358-ba47-d92df3d6a102 · inbound

Vibe Check: Understanding the Effects of LLM-Based Conversational Agents' Personality and Alignment on User Perceptions in Goal-Oriented Tasks cites this paper.

Vibe Check: Understanding the Effects of LLM-Based Conversational Agents' Personality and Alignment on User Perceptions in Goal-Oriented Tasks From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 114

Resolution: verified exact; arxiv_id, observed 2026-05-18T17:01:38.344182Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.

Observation 5896c4be-d5ba-4f1d-b9d4-3591bf8b6ef7 · inbound

Beyond Linear Probes: Dynamic Safety Monitoring for Language Models cites this paper.

Beyond Linear Probes: Dynamic Safety Monitoring for Language Models From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 63

Resolution: verified exact; arxiv_id, observed 2026-05-18T12:42:36.853762Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.

Observation 36ecda09-0937-4c16-ab76-00ee4d62cc54 · inbound

IatroBench: Pre-Registered Evidence of Iatrogenic Harm from AI Safety Measures cites this paper.

IatroBench: Pre-Registered Evidence of Iatrogenic Harm from AI Safety Measures From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 24

Resolution: verified exact; arxiv_id, observed 2026-05-11T00:35:52.616391Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.

Observation 07612579-4de2-4a67-a42e-7fd7c0e3ed24 · inbound

IatroBench: Pre-Registered Evidence of Iatrogenic Harm from AI Safety Measures cites this paper.

IatroBench: Pre-Registered Evidence of Iatrogenic Harm from AI Safety Measures From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 24

Resolution: unresolved; no resolver link, observed 2026-07-13T00:19:33.861692Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 66929448-40e1-4ddf-be66-fc7de901521d · inbound

Cat-DPO: Category-Adaptive Safety Alignment cites this paper.

Cat-DPO: Category-Adaptive Safety Alignment From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 25

Resolution: verified exact; arxiv_id, observed 2026-05-10T05:36:01.750925Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.

Observation 41e0bb4f-39d5-48a6-83fc-c31191e12e91 · inbound

Using large language models for embodied planning introduces systematic safety risks cites this paper.

Using large language models for embodied planning introduces systematic safety risks From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 83

Resolution: verified exact; arxiv_id, observed 2026-05-10T11:40:20.104188Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.

Observation b0bec346-0217-440a-9c94-502e42612725 · inbound

Jailbreaking Frontier Foundation Models Through Intention Deception cites this paper.

Jailbreaking Frontier Foundation Models Through Intention Deception From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 22

Resolution: verified exact; arxiv_id, observed 2026-05-11T22:11:14.010147Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.

Observation aec6227f-8076-41dc-94a9-9d0babd9efd4 · inbound

Chain of Risk: Safety Failures in Large Reasoning Models and Mitigation via Adaptive Multi-Principle Steering cites this paper.

Chain of Risk: Safety Failures in Large Reasoning Models and Mitigation via Adaptive Multi-Principle Steering From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 11

Resolution: verified exact; arxiv_id, observed 2026-05-11T19:31:08.563625Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.

Observation 27dddf4d-e254-4f60-953b-8ff3632f9d83 · inbound

Internalizing Safety Understanding in Large Reasoning Models via Verification cites this paper.

Internalizing Safety Understanding in Large Reasoning Models via Verification From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 26

Resolution: verified exact; arxiv_id, observed 2026-05-12T01:51:14.238142Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.

Observation 5d489190-aa42-498a-9084-49fc23f6703c · inbound

Reducing Political Manipulation with Consistency Training cites this paper.

Reducing Political Manipulation with Consistency Training From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 40

Resolution: verified exact; arxiv_id, observed 2026-05-22T05:34:40.233116Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.

Observation 27d5e496-392f-4511-a39f-bf7b49040453 · inbound

Reducing Political Manipulation with Consistency Training cites this paper.

Reducing Political Manipulation with Consistency Training From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 40

Resolution: verified exact; arxiv_id, observed 2026-06-30T16:54:58.727042Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.

Observation 1a845245-3a9c-46d1-9b9b-cdb20a841c83 · inbound

LLM Judges Inconsistently Disagree Across Safety Criteria and Harm Categories cites this paper.

LLM Judges Inconsistently Disagree Across Safety Criteria and Harm Categories From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 4

Resolution: metadata mismatch; arxiv_id, observed 2026-07-01T19:26:00.785841Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.

Observation d966b265-c9f3-4000-b32b-f8565878dee3 · inbound

Investigating and Alleviating Harm Amplification in LLM Interactions cites this paper.

Investigating and Alleviating Harm Amplification in LLM Interactions From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 24

Resolution: metadata mismatch; arxiv_id, observed 2026-07-01T23:16:24.040929Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.

Observation fed5df3e-33d3-415e-967a-4a6a64015832 · inbound

Safety Measurements for Fine-tuned LLMs Should be Grounded in Capability cites this paper.

Safety Measurements for Fine-tuned LLMs Should be Grounded in Capability From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 20

Resolution: metadata mismatch; arxiv_id, observed 2026-07-02T03:26:29.803972Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.

Observation 28e34803-b869-4ca7-93d5-9fb1e5c85b01 · inbound

Understanding Censorship in Large Language Models: From Mechanisms to Governance cites this paper.

Understanding Censorship in Large Language Models: From Mechanisms to Governance From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 42

Resolution: verified exact; arxiv_id, observed 2026-07-01T07:05:28.397555Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.

Observation 8032e4dd-60b3-47f1-a891-572cd3fd65ad · inbound

OpenSafeIntent: Evaluating Intent-Calibrated Safe Completion Across Dual-Use Prompt Sets cites this paper.

OpenSafeIntent: Evaluating Intent-Calibrated Safe Completion Across Dual-Use Prompt Sets From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 5

Resolution: metadata mismatch; arxiv_id, observed 2026-07-03T14:48:32.496890Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.

Observation 13e43977-b44e-4383-bfa6-66195151621c · inbound

Oyster-II: Reinforcement Learning for Constructive Safety Alignment in Large Language Models cites this paper.

Oyster-II: Reinforcement Learning for Constructive Safety Alignment in Large Language Models From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 62

Resolution: unresolved; no resolver link, observed 2026-07-12T06:11:38.403281Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation b3b09427-a1e8-4a06-a09e-bb0a28d19412 · inbound

GPT-Red: Automated Red Teaming via Self-Play at Scale cites this paper.

GPT-Red: Automated Red Teaming via Self-Play at Scale From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 15

Resolution: unresolved; no resolver link, observed 2026-08-01T01:12:44.184834Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3a138c4e-30e0-4b6a-8e2f-7bead74052f4 · inbound

Choosing Where and How to Moderate: End-to-End Trade-offs in Filter Placement and Response Rewriting cites this paper.

Choosing Where and How to Moderate: End-to-End Trade-offs in Filter Placement and Response Rewriting From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 20

Resolution: unresolved; no resolver link, observed 2026-08-01T00:35:58.478085Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9486a6b5-410d-471b-afd1-06341419d6b5 · inbound

Safeguards Based on Copyable Context Cannot Provide Reliable Safety for LLMs cites this paper.

Safeguards Based on Copyable Context Cannot Provide Reliable Safety for LLMs From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 9

Resolution: unresolved; no resolver link, observed 2026-07-31T22:25:22.224180Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation a2b2ff0b-4eeb-40cd-90bd-3899f92aa4c8 · inbound

Unaccountable Delegation, Fading Skills: Mapping the Risks of Workplace AI Agents cites this paper.

Unaccountable Delegation, Fading Skills: Mapping the Risks of Workplace AI Agents From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training

Reference 25

Resolution: unresolved; no resolver link, observed 2026-08-14T04:34:57.214509Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.