DeepSeek Gets an ‘F’ in Safety From Researchers
Cisco tested DeepSeek's open-source model, DeepSeek R1, which failed to fend off all 50 harmful behavior prompts from the HarmBench dataset. DeepSeek's failure rate is the highest among tested LLMs, with other models like Meta's Llama 3.1 and OpenAI's o1 performing noticeably better. The model's susceptibility to attacks, alongside data security concerns, has raised significant scrutiny and criticism.