AI Prompt Injection Test Suite: 122 Tests Across 11 Categories

A test suite of 122 prompt injection attacks across 11 categories designed to evaluate AI model resistance. It was built as part of the CyberRanger V42-Gold project by David Keane for an NCI MSc in Cybersecurity, referencing research from 2023. The dataset page was last updated on March 13, 2026.

Use Cases

• Benchmarking model resistance to prompt injection based on the 122-test battery.
• Evaluating jailbreak-resistant language models based on the described attack categories.
• Training or fine-tuning models for security hardening based on the test suite examples.

Strengths

• Contains 122 distinct test cases for systematic evaluation.
• Organized into 11 categories for structured analysis.
• Built with reference to specific academic research from 2023.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

DavidTKeane (David Keane), part of CyberRanger V42-Gold project for NCI MSc Cybersecurity.

Collection Method

Likely manually curated or synthesized from cited research (Greshake et al., Zou et al., Wei et al., 2023).

Freshness

Last updated 2026-03-13 02:17:20; freshness should be verified.