Anthropic CEO Dario Amodi, Chinese AI Company, is concerned about the competitive Deepsak To take Silicon Valley from the storm with Its R1 modelAnd their concerns can be more serious than specific people raised about the lampsac about sending user data to China.
In an interview On Jordan Schinider’s Chinatlak Podcast, Amodi said Deepsek generated rare information about biovipan in a safety test conducted by Anthropic.
Deepsek’s performance “was originally the worst of any model we had ever tested,” Amodi claimed. “It was no block against generating this information.”
Amodei stated that it was part of the evaluation, which runs regularly on various AI models to assess its potential national security risks. Their team sees whether models can generate bioweapons-related information that is not easily found on Google or in textbooks. Status as Anthropic AI Foundational Model Provider Which takes security seriously.
Amodi said that he did not think that the models of Deepsak are “literally dangerous” in providing rare and dangerous information today, but they could be in the near future. Although he praised the Deepsak’s team as “talented engineers”, he advised the company to “take these AI security views seriously”.
Amodei is also Supported strong export control On China’s chips, citing concerns that they can give an edge to the Chinese army.
Amodei did not clarify in the Chintok interview about which model anthropic testing was conducted, nor did they give more technical details about these tests. Anthropic immediately did not respond to a request to comment from Techcrunch. Neither Deepsek.
Uday of Deepsak has expressed concern about his safety elsewhere. For example, Cisco Security Researchers Said last week The Dipsek R1 failed to block any harmful signals in its safety tests, leading to 100% gelbreak success rate.
Cisco did not mention Biovipans, but said it was capable of obtaining Deepsek to generate harmful information about cyber crime and other illegal activities. However, it is worth noting that the GPT-4o of the Meta-3.1-405B and Openai also had a high failure rate of 96% and 86% respectively.
It remains to be seen whether such security concerns will make a serious dent in Deepsek AdoptLike companies AWS And Microsoft Publicly tied R1 to integrate into its cloud platform – irony is that Amazon is The largest investor of the anthropic,
On the other hand, Is a growing list Of countries, CompaniesAnd especially a government organization like US Navy and Pentagon It has started banning the lampsake.
Time will tell whether these efforts catch or if the global growth of Deepsak will continue. Either way, Amodi says that he consider Deepsek a new competitor that is at the level of the top AI companies in the US.
“The new fact here is that there is a new contestant,” he said on Chintok. “In big companies that AI – Anthropic, Open, Google, may probably train Meta and XE – now Deepsek is probably being added to that category.”
