Anthropic’s CEO Dario Amodei is worried about Deepseek competitors, a Chinese AI company that took Silicon Valley to the R1 model. And his concern can be more serious than the typical one raised about DEEPSEEK sending back to China.
In an interview with Jordan Schneider’s Chinatalk Podcast, Amodei said Deepseek generated rare information about Bioweapons in Anthropic’s safety test.
DeepSeek’s performance is “basically the worst of all the models we tested,” AMODEI claimed. “There was no block at all about creating this information.”
Amodei said this is part of the evaluation. His team examines whether the model can create information about Biowopons that can’t be easily found in Google or textbooks. Humanity is located in its own AI basic model provider that takes safety seriously.
Amodei said today that the model of Deepseek did not think it was “literally dangerous,” but it would be in the near future. He praised DeepSeek’s team as a “talented engineer,” but the company advised that he would take these AI safety considerations seriously.
AMODEI also mentioned concerns that it could take advantage of China’s military by supporting strong export control of China’s chips.
AMODEI did not clarify in an interview with the Interview with the DEEPSEEK model and did not provide more technical details on these tests. Anthropic did not immediately respond to TechCrunch’s request. The same was true for Deepseek.
DeepSeek’s Rise has raised concerns about safety elsewhere. For example, Cisco Security Researchers said last week that Deepseek R1 did not block harmful prompt in safety tests and achieved a 100% prisoner success rate.
CISCO said that it did not mention Biowopons, but could produce harmful information about cyber crimes and other illegal activities. However, it is worth mentioning that META’s LLAMA-3.1-405B and Openai’s GPT-4O have a high failure rate of 96%and 86%, respectively.
It still remains whether this safety problem will create a serious hollow place for the quick adoption of DeepSeek. Companies such as AWS and Microsoft have publicly promoted the integration of R1 into cloud platforms. It is ironically enough to consider that Amazon is the largest investor in Anthropic.
On the other hand, the list of government agencies such as the state, the company, especially the US Navy and the Ministry of Defense, is growing.
It will tell you whether this effort will continue to rise in the global rise of Deepseek. Either way, Amodei thinks DeepSeek is a new competitor at the level of the best AI company in the United States.
“The new fact is that there is a new competitor,” China said. “In a large company that can train AI (Anthropic, Openai, Google, Perhaps Meta and XAI), DeepSeek will be added to the category.”
