SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2601.04740 · 2026

StealthGraph: Exposing Domain-Specific Risks in LLMs through Knowledge-Graph-Guided Harmful Prompt Generation

Sen Yang, Shouling Ji, Dazhen Deng, Yingcai Wu, Huawei Zheng, Xinqi Jiang

arXiv · PDF · Open in the Atlas

Code that ran

4 of 4 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, 0 violated it and 0 have no contract verdict; 3 failures of our instrument; 0 no recorded run), measured 2026-11-22; 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/2601.04740.json · badge: /api/ran/2601.04740.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
ZJUIDG-AIVA/StealthGraph canonical (classifier label)
classifier basis: census
4 ran · 0 no recorded run
FunctionStatusWhere it lives
compute_toxicity_score Ran
ran, honoured the contract we drafted
ZJUIDG-AIVA/StealthGraph/experiment/exp4_ablation_generation_model_and_iteration_and_dual_path.py
code served (permissive licence) · get_code("4f5ba97f8774df40")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import json
import tempfile
import os

# Create a temporary JSON file with toxicity evaluation data
temp_dir = tempfile.mkdtemp()
json_path = os.path.join(temp_dir, 'toxicity_data.json')

# Create sample data matching the function's expected structure
sample_data = [
    {
        'toxicity_evaluation': {
            'toxicity_score': 0.25
        }
    },
    {
        'toxicity_evaluation': {
            'toxicity_score': 0.45
        }
    },
    {
        'toxicity_evaluation': {
            'toxicity_score': 0.30
        }
    }
]

# Write the sample data to the temporary JSON file
with open(json_path, 'w', encoding='utf-8') as f:
    json.dump(sample_data, f)

ARGS = {
    'path': json_path,
    'json_key': 'toxicity_evaluation'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
read_steal_success_texts Ran
ran; our contract draft was wrong, not the code
ZJUIDG-AIVA/StealthGraph/experiment/exp4_ablation_generation_model_and_iteration_and_dual_path.py
code served (permissive licence) · get_code("51bef3f6aa2252cd")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import json
import tempfile
import os

# Create a temporary JSON file with valid test data
temp_dir = tempfile.mkdtemp()
test_json_path = os.path.join(temp_dir, 'test_stealth.json')

# Create test data with stealth_success entries
test_data = [
    {
        'stealth_success': True,
        'stealth_prompt': 'This is the first stealth prompt text'
    },
    {
        'stealth_success': True,
        'stealth_prompt': 'This is the second stealth prompt text'
    },
    {
        'stealth_success': False,
        'stealth_prompt': 'This should not be included'
    },
    {
        'stealth_success': True,
        'stealth_prompt': 'This is the third stealth prompt text'
    }
]

# Write test data to temporary JSON file
with open(test_json_path, 'w', encoding='utf-8') as f:
    json.dump(test_data, f)

ARGS = {
    'path': test_json_path,
    'json_key': 'stealth_prompt'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
read_texts Ran
ran; our contract draft was wrong, not the code
ZJUIDG-AIVA/StealthGraph/experiment/exp4_ablation_generation_model_and_iteration_and_dual_path.py
code served (permissive licence) · get_code("cbdffc507bdcf6dd")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import json
import tempfile
import os

# Create a temporary JSON file with test data
temp_dir = tempfile.mkdtemp()
test_json_path = os.path.join(temp_dir, 'test_data.json')

test_data = [
    {'prompt_text': 'This is the first text sample.'},
    {'prompt_text': 'This is the second text sample.'},
    {'prompt_text': 'This is the third text sample.'},
]

with open(test_json_path, 'w', encoding='utf-8') as f:
    json.dump(test_data, f)

ARGS = {
    'path': test_json_path,
    'json_key': 'prompt_text'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
read_texts Ran
ran; our contract draft was wrong, not the code
ZJUIDG-AIVA/StealthGraph/experiment/exp4_ablation_kg.py
code served (permissive licence) · get_code("cc0ac0a505e201f0")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import json
import tempfile
import os

# Create a temporary JSON file with test data
temp_dir = tempfile.mkdtemp()
json_path = os.path.join(temp_dir, 'test_data.json')

test_data = [
    {'prompt_text': 'This is the first text sample.'},
    {'prompt_text': 'This is the second text sample.'},
    {'prompt_text': 'This is the third text sample.'},
]

with open(json_path, 'w', encoding='utf-8') as f:
    json.dump(test_data, f)

ARGS = {
    'path': json_path,
    'json_key': 'prompt_text'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block

Also present in other papers' code

Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 3 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 3 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.

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

Large language models (LLMs) are increasingly applied in specialized domains such as finance and healthcare, where they introduce unique safety risks. Domain-specific datasets of harmful prompts remain scarce and still largely rely on manual construction; public datasets mainly focus on explicit harmful prompts, which modern LLM defenses can often detect and refuse. In contrast, implicit harmful prompts-expressed through indirect domain knowledge-are harder to detect and better reflect real-world threats. We identify two challenges: transforming domain knowledge into actionable constraints and increasing the implicitness of generated harmful prompts. To address them, we propose an end-to-end framework that first performs knowledge-graphguided harmful prompt generation to systematically produce domain-relevant prompts, and then applies two-strategy obfuscation rewriting to convert explicit harmful prompts into implicit variants via direct and context-enhanced rewriting. This framework yields high-quality datasets combining strong domain relevance with implicitness, enabling more realistic redteaming and advancing LLM safety research. We release our code and datasets on GitHub. 1

For agents

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

get_harvested_code_for_paper("2601.04740")
get_code_for_paper("2601.04740")
have("2601.04740")

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

curl https://syntology.ai/api/ran/2601.04740.json

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

[![Syntology run record](https://syntology.ai/api/ran/2601.04740.svg)](https://syntology.ai/paper/2601.04740)

Connect an agent — have() is free.