DataBoundary: A Benchmark for Delimiter-Based Prompt Injection Defense Across 13 LLMs

DataBoundary is a red/blue team benchmark dataset evaluating delimiter-based defenses against prompt injection attacks. It contains 5,578 test cases, testing 3 attack templates and 7 injection payloads across 13 LLMs, including cloud APIs and locally-hosted models. The dataset was created by Alan-StratCraftsAI and was last updated on 2026-05-05.

Use Cases

• Benchmarking the effectiveness of delimiter-based boundary markers against prompt injection based on the described defense mechanism.
• Comparing the vulnerability of different LLMs to injection attacks based on the 13 models tested.
• Developing new prompt injection attack templates based on the 3 existing templates in the dataset.
• Evaluating the robustness of LLM-integrated applications against adversarial payloads based on the 7 injection payloads.

Strengths

• Contains 5,578 test cases for systematic evaluation.
• Evaluates defenses across 13 different LLMs, providing broad model coverage.
• Includes 3 distinct attack templates and 7 injection payloads for varied testing scenarios.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last updated 2026-05-05 00:33:53; freshness should be verified.

Provenance

Source

Alan-StratCraftsAI via Hugging Face

Collection Method

Likely generated through red/blue team adversarial testing.

Time Range

Benchmark evaluation period from 2026-04-30 to 2026-05-04

Freshness

2026-05-05 00:33:53