Adversarial Robustness of LLM-Based Multi-Agent Systems for Engineering Problems

One of the first systematic studies of adversarial robustness in LLM-based multi-agent systems applied to engineering contexts. The research investigates error propagation under controlled adversarial influence across problems like pipe pressure loss, beam deflection, and graph traversal. Authored by Lorenz Wiesmeier and shared under a CC-BY-4.0 license on figshare in May 2026.

Use Cases

• Benchmarking adversarial robustness of LLM-based agents based on task type and communication order described in the study
• Designing resilient multi-agent system architectures based on findings about prompt framing and agent role assignment
• Studying error propagation in collaborative AI reasoning based on experiments with misleading agents
• Evaluating domain-specific adversarial vulnerabilities based on tests with structurally complex engineering tasks

Strengths

• The study is one of the first systematic investigations of adversarial robustness in LLM-based MAS for engineering, as stated in the description.
• The research quantifies error propagation across four representative engineering problems, providing concrete test cases.
• The dataset is shared under a permissive CC-BY-4.0 license, facilitating reuse.

Limitations

• The dataset is a 1.9 MB PDF file, which is a small-scale research paper rather than a structured data corpus.
• Column-level documentation and sample data are unavailable, limiting direct data analysis.
• Row count is unknown, which may limit suitability assessment for quantitative modeling.

Provenance

Source

figshare

Collection Method

Likely contains research results and analysis from systematic adversarial testing.

Freshness

Last updated 2026-05-13 05:51:48; freshness should be verified.