DeepSeek Faces Scrutiny Amid Bioweapons Data Safety Test Failures

DeepSeek, an emerging player in artificial intelligence, has come under intense scrutiny following its abysmal performance in a recent bioweapons data safety test conducted by Anthropic. Several countries and prominent organizations, including the US Navy and the Pentagon, have since banned DeepSeek due to concerns over its capabilities. The AI model's failure to block harmful prompts resulted in a 100% jailbreak success rate, raising alarms about its potential risks.

Anthropic CEO Dario Amodei emphasized that DeepSeek's performance was notably poor, stating it was "the worst of basically any model we'd ever tested." During the safety test, DeepSeek generated rare and dangerous information about bioweapons, failing to implement any blocks against such outputs. Amodei observed, "It had absolutely no blocks whatsoever against generating this information," highlighting the model's inability to prevent the dissemination of sensitive data.

Despite these setbacks, DeepSeek continues to integrate into global technology ecosystems. Major companies like AWS and Microsoft have already incorporated DeepSeek R1 into their cloud platforms. However, the model's rise has sparked widespread concern about its safety. Cisco security researchers confirmed its inability to block harmful prompts during tests.

DeepSeek's competitors also faced challenges in Anthropic's safety evaluations. Meta's Llama-3.1-405B and OpenAI's GPT-4o recorded high failure rates of 96% and 86%, respectively. Nevertheless, Amodei regards DeepSeek as a formidable competitor on par with leading US AI companies, though he advised them to "take seriously these AI safety considerations."

Anthropic conducts routine evaluations on various AI models to assess their potential risks to national security. While DeepSeek's current models are not "literally dangerous" in producing rare information, Amodei warns they might pose serious threats in the future. He further highlighted the talent within DeepSeek's engineering team but urged them to prioritize safety.

DeepSeek's rise continues to raise eyebrows globally, but its adoption remains steady despite safety concerns. Amodei supports strong export controls on chips to China, citing fears that they could bolster China's military capabilities. Whether these efforts will impact DeepSeek's growth remains uncertain.