Anthropic CEO says Deepseek is a worse thing for ‘Biowepons security testing

Anthshopic’s CEO Dario Amode are concerned about the form of competition, China Ai taking Silicon Valley in a storm with its R1 model. And his anxiety can be very bad than usual nominated by sending user data back to China.

In the Jordan Schneider Podcast Podcast, Amodei said Deepseeek produced unverified information about bioweathons in the security test driven by anthropic.

Deepseeek function was “the worst of the basics of any model that has been examined,” said Amodii. “There are no totally symptoms facing this information.”

Amodei said that this was part of anthropic anthropic anthromism that runs various models in AI to examine their potential security risks. His team looks at models that can produce biowewons related information that is not easily available on Google or books. Anthropic positions itself as a Basic AI model of AI who takes a deep security.

Amodeii said he did not think that Seviseeeek Models were “realistic” in gained and dangerous information but talented. “

Amodei also supports solid shipping controls in China China, quoting anxiety can give Chinese soldiers.

Amodei did not specify Chinatal Composition which is in the defective Anthropic model assessed, and did not provide multiple technical information about these tests. Anthropic did not respond immediately to the comment request from Techcruk. And no problem.

SeppseEek’s Heights Revurn the anxiety of its safety elsewhere, too. For example, Cisco security researchers said last week that Deepseek R1 failed to prevent any harmful rates for the security test, reaching the 100% success rate.

Cisco didn’t mean bioweathons but he said he could get deep productivity to produce harmful information about cybercrime and illegal activities. It is appropriate to say, however, that Llama’s Llama – 3.1-405b and Open-4o also has higher 96% failure and 86%, respectively.

It is yet to be seen that worrying about such security will make it difficult for the speedy gathing of Deepseeek. Companies such as Microsoft has been discharged from the public including R1 in their platforms – unfortunately sufficient, provided that Amazon is anthropic investor.

On the other hand, there are the growing list of countries, companies, and government organizations, especially the US Navy and Pentagon starting to block Deentieseek.

The time will mean that these efforts are hosting or when the global population world will continue. Any way, Amodei says that he consider the depth of a new competition of the highest AI.

“The new truth here is that there is a new competition,” says Chinatalk. “In large companies that can train AI – Anthropic, an Acabo, Google, maybe Meta and Xai – Now the Deepseek is probably added to that section.”