Are LLM-Enhanced Graph Neural Networks Robust against Poisoning Attacks? | Signal Canvas | ScienceToStartup

Are LLM-Enhanced Graph Neural Networks Robust against Poisoning Attacks? | Signal Canvas | ScienceToStartup

Page Freshness

Signal Canvas proof surface

Canonical route: /signal-canvas/are-llm-enhanced-graph-neural-networks-robust-against-poisoning-attacks

stale

Proof freshness: stale
Proof status: partial
Display score: 7/10
Last proof check: 2026-03-30
Score updated: 2026-04-02
Score fresh until: 2026-05-02
References: 64
Source count: 4
Coverage: 83%

This page is showing the last landed evidence receipt and score bundle because the latest proof data is outside the freshness window.

Agent Handoff

Are LLM-Enhanced Graph Neural Networks Robust against Poisoning Attacks?

Canonical ID are-llm-enhanced-graph-neural-networks-robust-against-poisoning-attacks | Route /signal-canvas/are-llm-enhanced-graph-neural-networks-robust-against-poisoning-attacks

REST example

curl https://sciencetostartup.com/api/v1/agent-handoff/signal-canvas/are-llm-enhanced-graph-neural-networks-robust-against-poisoning-attacks

MCP example

{
  "tool": "search_signal_canvas",
  "arguments": {
    "mode": "paper",
    "paper_ref": "are-llm-enhanced-graph-neural-networks-robust-against-poisoning-attacks",
    "query_text": "Summarize Are LLM-Enhanced Graph Neural Networks Robust against Poisoning Attacks?"
  }
}

source_context

{
  "surface": "signal_canvas",
  "mode": "paper",
  "query": "Are LLM-Enhanced Graph Neural Networks Robust against Poisoning Attacks?",
  "normalized_query": "2603.26105",
  "route": "/signal-canvas/are-llm-enhanced-graph-neural-networks-robust-against-poisoning-attacks",
  "paper_ref": "are-llm-enhanced-graph-neural-networks-robust-against-poisoning-attacks",
  "topic_slug": null,
  "benchmark_ref": null,
  "dataset_ref": null
}

Evidence Receipt

Route status: building

Claims: 12

References: 64

Proof: Verification pending

Freshness state: computing

Source paper: Are LLM-Enhanced Graph Neural Networks Robust against Poisoning Attacks?

PDF: https://arxiv.org/pdf/2603.26105v1

Repository: https://github.com/CyberAlSec/LLMEGNNRP}

Source count: 4

Coverage: 83%

Last proof check: 2026-03-30T20:30:36.673Z

Signal Canvas receipt window

Ready for execution: Are LLM-Enhanced Graph Neural Networks Robust against Poisoning Attacks?

/buildability/are-llm-enhanced-graph-neural-networks-robust-against-poisoning-attacks

Build Nowready

Subject: Are LLM-Enhanced Graph Neural Networks Robust against Poisoning Attacks?

Verdict

Build Now

Verdict is Build Now because viability and implementation proof cleared the Wave 1 scaffold thresholds.

GitHub Code Pulse

No public code linked for this paper yet.

Claim map

Strong 12Mixed 0Weak 0

Evidencepartial
To bridge this gap, we propose a robustness assessment framework that systematically evaluates LLM-enhanced GNNs under poisoning attacks.
Implicationpartial
The abstract explicitly states the proposal of this framework to address the research gap.
Verificationpartial
partial
Evidencepartial
Specifically, we assess 24 victim models by combining eight LLM- or Language Model (LM)-based feature enhancers with three representative GNN backbones.
Implicationpartial
The abstract and the parsed sections clearly detail the number of models and their composition.
Verificationpartial
partial
Evidencepartial
To ensure diversity in attack coverage, we incorporate six structural poisoning attacks (both targeted and non-targeted) and three textual poisoning attacks operating at the character, word, and sentence levels.
Implicationpartial
The abstract and parsed sections provide specific numbers and types of attacks used for diversity.
Verificationpartial
partial
Evidencepartial
Furthermore, we employ four real-world datasets, including one released after the emergence of LLMs, to avoid potential ground truth leakage during LLM pretraining, thereby ensuring fair evaluation.
Implicationpartial
The abstract and parsed sections explicitly mention the use of four datasets and the rationale for including a post-LLM dataset.
Verificationpartial
partial
Evidencepartial
Extensive experiments show that LLM-enhanced GNNs exhibit significantly higher accuracy and lower Relative Drop in Accuracy (RDA) than a shallow embedding-based baseline across various attack settings.
Implicationpartial
The abstract directly states this experimental finding.
Verificationpartial
partial
Evidencepartial
Our in-depth analysis identifies key factors that contribute to this robustness, such as the effective encoding of structural and label information in node representations.
Implicationpartial
The abstract mentions that in-depth analysis identifies these key factors contributing to robustness.
Verificationpartial
partial
Evidencepartial
To the best of our knowledge, this framework provides the most comprehensive coverage of victim models and attack types to date, and incorporates post-LLM dataset evaluation.
Implicationpartial
The paper explicitly claims this comprehensiveness in its contribution summary.
Verificationpartial
partial
Evidencepartial
To bridge this gap, we propose a robustness assessment framework that systematically evaluates LLM-enhanced GNNs under poisoning attacks.
Implicationpartial
The abstract explicitly states the proposal of this framework to address the research gap.
Verificationpartial
partial
Evidencepartial
Specifically, we assess 24 victim models by combining eight LLM- or Language Model (LM)-based feature enhancers with three representative GNN backbones.
Implicationpartial
The abstract provides specific numbers for the models and components evaluated within the framework.
Verificationpartial
partial
Evidencepartial
To ensure diversity in attack coverage, we incorporate six structural poisoning attacks (both targeted and non-targeted) and three textual poisoning attacks operating at the character, word, and sentence levels.
Implicationpartial
The abstract details the types and granularities of the poisoning attacks used in the evaluation.
Verificationpartial
partial
Evidencepartial
Furthermore, we employ four real-world datasets, including one released after the emergence of LLMs, to avoid potential ground truth leakage during LLM pretraining, thereby ensuring fair evaluation.
Implicationpartial
The abstract specifies the number and type of datasets used, highlighting the rationale for including a post-LLM dataset.
Verificationpartial
partial
Evidencepartial
Extensive experiments show that LLM-enhanced GNNs exhibit significantly higher accuracy and lower Relative Drop in Accuracy (RDA) than a shallow embedding-based baseline across various attack settings.
Implicationpartial
The abstract directly states the experimental findings comparing LLM-enhanced GNNs to a baseline.
Verificationpartial
partial

Author intelligence and commercialization panels stay hidden until the proof receipt is verified, cites at least 3 references, includes at least 2 sources, and clears 50% coverage. The paper narrative and citation surfaces remain public while verification is pending.