SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2404.01099 · 2024

What is in Your Safe Data? Identifying Benign Data that Breaks Safety

arXiv · PDF · Open in the Atlas

Code that ran

7 of 14 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 5 failures of our instrument; 7 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2404.01099.json · badge: /api/ran/2404.01099.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
princeton-nlp/benign-data-breaks-safety canonical (classifier label)
classifier basis: census
7 ran · 7 no recorded run
FunctionStatusWhere it lives
contains_unwanted_words Ran
ran, violated the contract we drafted
princeton-nlp/benign-data-breaks-safety/ft_datasets/alpaca_dataset/alpaca_remove_safety.py
pointer only (licence: NONE) · get_code("c4a58b263cc3b3d7")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, princeton-nlp/benign-data-breaks-safety when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

extract_content Ran
ran, honoured the contract we drafted
princeton-nlp/benign-data-breaks-safety/safety_evaluation/eval_utils/openai_gpt4_judge.py
pointer only (licence: NONE) · get_code("e28da1bf4c00b938")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, princeton-nlp/benign-data-breaks-safety when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_credentials Ran
ran; our contract draft was wrong, not the code
princeton-nlp/benign-data-breaks-safety/safety_evaluation/eval_utils/api_credentials.py
pointer only (licence: NONE) · get_code("f8f41288d948ed1a")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, princeton-nlp/benign-data-breaks-safety when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

load_llama_from_config Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
princeton-nlp/benign-data-breaks-safety/safety_evaluation/eval_utils/model_utils.py
pointer only (licence: NONE); the stored copy get_code serves, from zhu-minjun/safetylock, is pointer only (licence: NONE) · get_code("b4c052fb373aef07")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, princeton-nlp/benign-data-breaks-safety when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

load_peft_model Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
princeton-nlp/benign-data-breaks-safety/safety_evaluation/eval_utils/model_utils.py
pointer only (licence: NONE); the stored copy get_code serves, from zhu-minjun/safetylock, is pointer only (licence: NONE) · get_code("6d5ba8546b6aad2c")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, princeton-nlp/benign-data-breaks-safety when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

question_read Ran
ran; our contract draft was wrong, not the code
princeton-nlp/benign-data-breaks-safety/safety_evaluation/question_inference.py
pointer only (licence: NONE) · get_code("4b6ac9bdbe621603")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, princeton-nlp/benign-data-breaks-safety when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

read_dialogs_from_file Ran
ran; our contract draft was wrong, not the code
princeton-nlp/benign-data-breaks-safety/safety_evaluation/eval_utils/chat_utils.py
pointer only (licence: NONE); the stored copy get_code serves, from yxduir/LLM-SRT, is pointer only (licence: NONE) · get_code("8922fb4b5b7006f7")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, princeton-nlp/benign-data-breaks-safety when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

contains_unwanted_words Not yet run
unverified (harvested, no recorded run)
princeton-nlp/benign-data-breaks-safety/ft_datasets/dolly_dataset/dolly_remove_safety.py
pointer only (licence: NONE) · get_code("345f0b1ebadc1536")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, princeton-nlp/benign-data-breaks-safety when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

duo_judge Not yet run
unverified (harvested, no recorded run)
princeton-nlp/benign-data-breaks-safety/safety_evaluation/eval_utils/openai_gpt4_judge.py
pointer only (licence: NONE) · get_code("30139d8373ac2f31")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, princeton-nlp/benign-data-breaks-safety when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

duo_judge_v2 Not yet run
unverified (harvested, no recorded run)
princeton-nlp/benign-data-breaks-safety/safety_evaluation/eval_utils/openai_gpt4_judge.py
pointer only (licence: NONE) · get_code("c77c19021491eb81")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, princeton-nlp/benign-data-breaks-safety when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

format_tokens Not yet run
unverified (harvested, no recorded run)
princeton-nlp/benign-data-breaks-safety/safety_evaluation/eval_utils/chat_utils.py
pointer only (licence: NONE); the stored copy get_code serves, from zhu-minjun/safetylock, is pointer only (licence: NONE) · get_code("e084e801c0e634ec")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, princeton-nlp/benign-data-breaks-safety when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

load_model Not yet run
unverified (harvested, no recorded run)
princeton-nlp/benign-data-breaks-safety/safety_evaluation/eval_utils/model_utils.py
pointer only (licence: NONE); the stored copy get_code serves, from zhu-minjun/safetylock, is pointer only (licence: NONE) · get_code("efea256714a4cb99")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, princeton-nlp/benign-data-breaks-safety when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

skip Not yet run
unverified (harvested, no recorded run)
princeton-nlp/benign-data-breaks-safety/ft_datasets/alpaca_dataset/alpaca_remove_safety.py
pointer only (licence: NONE) · get_code("84cf7fe41466b4b3")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, princeton-nlp/benign-data-breaks-safety when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

skip Not yet run
unverified (harvested, no recorded run)
princeton-nlp/benign-data-breaks-safety/ft_datasets/dolly_dataset/dolly_remove_safety.py
pointer only (licence: NONE) · get_code("a01a8759063203f2")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, princeton-nlp/benign-data-breaks-safety when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 14 distinct functions with an identity, 6 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 0 the same month, 6 later) and 8 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
load_peft_model 7 other papers
5 dated earlier, 2 later
2308.10855 (2023-08)
LatEval: An Interactive LLMs Evaluation Benchmark with Incomplete Information from Lateral Thinking Puzzles
2310.02469 (2023-10)
PrivacyMind: Large Language Models Can Be Contextual Privacy Protection Learners
2310.06498 (2023-10)
A New Benchmark and Reverse Validation Method for Passage-level Hallucination Detection
and 4 more not named here
load_llama_from_config 6 other papers
4 dated earlier, 2 later
2310.02469 (2023-10)
PrivacyMind: Large Language Models Can Be Contextual Privacy Protection Learners
2310.06498 (2023-10)
A New Benchmark and Reverse Validation Method for Passage-level Hallucination Detection
2310.08491 (2023-10)
Prometheus: Inducing Fine-grained Evaluation Capability in Language Models
and 3 more not named here
load_model 3 other papers
2 dated earlier, 1 later
2308.10855 (2023-08)
LatEval: An Interactive LLMs Evaluation Benchmark with Incomplete Information from Lateral Thinking Puzzles
2310.02469 (2023-10)
PrivacyMind: Large Language Models Can Be Contextual Privacy Protection Learners
2410.10343 (2024-10)
Locking Down the Finetuned LLMs Safety
read_dialogs_from_file 3 other papers
3 later
2410.10343 (2024-10)
Locking Down the Finetuned LLMs Safety
2025.findings-acl.337 (2025 (year only))
The Self-Improvement Paradox: Can Language Models Bootstrap Reasoning Capabilities without External Scaffolding?
2602.21646 (2026-02)
Scalable Multilingual Multimodal Machine Translation with Speech-Text Fusion
format_tokens 2 other papers
2 later
2410.10343 (2024-10)
Locking Down the Finetuned LLMs Safety
2602.21646 (2026-02)
Scalable Multilingual Multimodal Machine Translation with Speech-Text Fusion
get_credentials 1 other paper
1 later
2406.17055 (2024-06)
Large Language Models Assume People are More Rational than We Really are

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Current Large Language Models (LLMs), even those tuned for safety and alignment, are susceptible to jailbreaking. Some have found that just further fine-tuning an aligned model with benign data (i.e., data without harmful content) surprisingly leads to substantial degradation in safety. We delve into the data-centric aspects of why benign fine-tuning inadvertently contributes to jailbreaking. First, we represent fine-tuning data through two lenses: representation and gradient spaces. Additionally, we propose a bi-directional anchoring method that, during the selection process, prioritizes data points that are close to harmful examples and far from benign ones. Our approach effectively identifies subsets of benign data that are more likely to degrade the model's safety after fine-tuning. Training on just 100 of these seemingly benign datapoints surprisingly leads to the fine-tuned model affirmatively responding to >70% of tested harmful requests, compared to <20% after fine-tuning on randomly selected data. We also observe that the selected data frequently appear as lists, bullet points, or math questions, indicating a systematic pattern in fine-tuning data that contributes to jailbreaking.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2404.01099")
get_code_for_paper("2404.01099")
have("2404.01099")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2404.01099.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2404.01099.svg)](https://syntology.ai/paper/2404.01099)

Connect an agent — have() is free.