This research presents a systematic bilingual (Chinese-English) safety evaluation of DeepSeek's large language and multimodal models, identifying potential risks and safety boundaries.

• Evaluates safety across multiple model types (text-only LLMs, multimodal LLMs, and text-to-image models)
• Develops a bilingual safety dataset specifically accounting for Chinese sociocultural contexts
• Identifies specific vulnerability patterns and safety limitations within these models
• Provides a methodological framework for comprehensive AI safety assessment

This research is critical for security professionals as it establishes systematic evaluation approaches for AI safety, helping organizations better understand, mitigate, and communicate potential risks when deploying frontier AI systems.

Towards Understanding the Safety Boundaries of DeepSeek Models: Evaluation and Findings