According to the anthropic CEO, Deepseek was “the worst” for the critical Bioupons data security test ncvrs.com

Anthropic CEO Dario Amodei is concerned about Deepseek competitor, China’s AI company, which carried the Silicon Valley with a storm with the R1 model. And his concerns can be more serious than typically suggested that the depth of user data sent him to China.

-In an interview At the Jordan Schneider Chinatat podcast, Amodei said Deepseek generated rare information about biophemes in the anthropic security test.

Deepseek lecture “The worst basically any model we have ever tested,” said Amodei. “There was no block against this information.”

Amodei stated that this is part of anthropic routine on various AI models of anthropic evaluations to assess their potential national security risks. Your team investigates whether models can generate bioapons information that is not easily found in Google or textbooks. Antropic is located as a service provider for AI basic model This takes security seriously.

Amodei said he did not think that Deepseek models are “literally dangerous” today when they provide rare and dangerous information, but that they could be in the near future. Although he praised the Deepseek team as “talented engineers”, he advised the company to “take these security considerations seriously.”

Amodei also supported China’s strong export controls, citing concerns that they could give concern to China’s military.

Amodei did not clarify in Chinatalk’s interview, which Deepseek tested the anthropic model and gave no technical details of these tests. The anthropic did not answer immediately at the request of Techcrunch note. Nothing Deepseek.

The rise of Deepseek has caused concern elsewhere in its security. For example, Cisco’s security researchers said last week This Deepseek R1 could not block harmful instructions in its security tests, reaching the 100% prison success rate.

Cisco did not mention Bioupons, but said it was capable of generating depth information about computer crime and other illegal activities. It is worth mentioning, however, that Meta Llama-3.1-405B and Openai GPT-4O are also high in failure 96%and 86%respectively.

You do not yet have to see that such security concerns are causing a serious hook in the rapid adoption of eea deepseek. Companies like Aws And Microsoft publicly integrated R1 integration into cloud platforms – ironically, as Amazon is the largest investor in the anthropic.

On the other hand, countries, companies and government organizations, such as the US Navy and Pentagon, are growing, and they have begun banning Deepsees.

Time tells you that these efforts will accelerate or if the global rise in Deepseek continues. Either way, Amodei says he considers Deepseek as a new competitor, which is at the level of the most popular AI companies in the United States.

“The new fact here is that there is a new competitor,” Chinataton said. “In large companies that can train AI -anthropic, Openai, Google, maybe Meta and Xai -now Deepseek is added to this category.”