Nemotron Content VISafe V1: 3,212 Vietnamese AI Safety Probes

3,212 text probes in Vietnamese for evaluating AI model safety across categories like jailbreak, toxicity, misinformation, and Vietnam-specific political sensitivity. The dataset combines translated probes from established English benchmarks with Vietnamese-native content and was created by NVIDIA. The dataset page was last updated on 2026-06-30.

Use Cases

• Benchmarking model refusal behavior based on jailbreak and prompt injection probes mentioned in the description
• Evaluating toxicity detection systems based on Vietnamese-language toxicity probes
• Testing for over-refusal or under-refusal based on the safety-critical prompt categories
• Assessing model performance on regionally sensitive topics based on Vietnam-specific political probes

Strengths

• Contains 3,212 validated text probes across multiple safety categories.
• Includes Vietnam-specific political sensitivity probes, offering regional relevance.
• Combines translated content from established benchmarks with native Vietnamese probes.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known but other scale details like file formats and size are unknown.

Provenance

Source

NVIDIA via Hugging Face.

Collection Method

Combines translated probes from established English safety benchmarks with Vietnamese-native content.

Freshness

Last updated 2026-06-30 11:18:13; freshness should be verified.

Geography

Focus on Vietnamese language and Vietnam-specific content.