Turkish Llm Injection

Turkey's first Turkish-priority, categorized LLM prompt injection dataset. It contains 300 manually and generation-assisted prepared Turkish prompt injection payloads, mapped to 12 attack categories and the OWASP LLM Top 10 (2025). The dataset was created by AltaySec and last updated on June 13, 2026.

Use Cases

• Training and evaluating LLM guardrails based on the 300 Turkish prompt injection payloads.
• Benchmarking model vulnerability across the 12 defined attack categories.
• Developing detection systems for adversarial prompts targeting Turkish-language models.
• Studying the characteristics of prompt injection attacks aligned with the OWASP LLM Top 10 framework.

Strengths

• Contains 300 specifically prepared Turkish prompt injection payloads.
• Payloads are categorized into 12 distinct attack types and mapped to the OWASP LLM Top 10 (2025).
• Version 0.2 adds 180 new payloads based on findings from an adversarial arena, suggesting iterative quality improvement.

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

AltaySec

Collection Method

Payloads were manually and generation-assisted prepared, with new additions based on findings from the AltayDuel adversarial arena.

Freshness

Last updated 2026-06-13 14:39:34; freshness should be verified.

Geography

Turkey (Turkish-language focus)