Anthropic Director General Dario Amodei is concerned about Deepseek competition, the Chinese company that received Silicon Valley from the storm with its R1 model. And his concerns can be more serious than the typical ones raised in connection with Deepseek Sending user data back to China.
In an interview in Jordan Schneider’s Podcast Chinatalk, Amodei said Deepseek created rare information about Bioweapons in an anthropic -led security test.
Deepseek’s performance was “the worst from essentially any model we would ever test,” Amodei claimed. “She had absolutely no blocks against generating this information.”
Amodei stated that this was part of anthropic estimates routinely in different models of him to assess their possible risks of national security. His team looks at whether models can generate bioapons related information that is not easily found on Google or textbooks. Anthropic positions itself as the provider of the basic model of one who takes security seriously.
Amodei said he did not think that Deepseek’s models today are “literally dangerous” in providing rare and dangerous information, but that they may be in the near future. Although he praised Deepseek’s team as “talented engineers”, he advised the company to “take these security considerations seriously”.
Amodei has also supported strong export controls on chips in China, citing concerns that they could give China’s army an advantage.
Amodei did not explain in the chinatalk interview which Deepseek model was tested anthropic, nor did he provide more technical details about these tests. Anthropic did not immediately respond to a comment request from Techcrunch. Nor Deepseek.
Deepseek’s growth has sparked concerns about its security elsewhere, too. For example, Cisco Security researchers said last week that the Deepseek R1 failed to block any harmful requests in his security tests, reaching a 100% prison success rate.
Cisco did not mention Boweapons but said he was able to get Deepseek to generate harmful information about internet crime and other illegal activities. It is worth noting, however, that Meta’s Llama-3.1-405b and OpenAi’s GPT-4o also had high rates of 96% and 86%, respectively.
It remains to be seen whether security concerns like these will make a serious trace in Deepseek’s rapid adoption. Companies like AWS and Microsoft have publicly defended the integration of R1 into their Cloud platforms – quite ironically, given that Amazon is the largest anthropic investor.
On the other hand, there is a growing list of countries, companies and especially government organizations such as the US Navy and the Pentagon that have begun to stop Deepseek.
Time will tell if these efforts capture or if the global rise of the Deepseek will continue. Either way, Amodei says he considers Deepseek a new competitor who is at the level of the best US companies.
“The new fact here is that there is a new competitor,” he told Chinatalk. “In the big companies he can train – anthropics, Openai, Google, probably Meta and Xai – now Deepseek is probably being added to that category.”
