Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Anthropic CEO Dario Amodani is worried about the opponent DeepSeek, the Chinese company Silicone took the valley with a storm with its R1 model. And his concerns can be more serious than the typical ones that Deepseek lifted the user information back to China.
In an interview Jordan Schneider’s Chinatalk Podcast is Amodei, DeepSeeteTeeton’s anthropic security tested in the security test that has created rare information about Bioweapons, he said.
Deepseek’s performance was “most worst of any model we tested.” “There was no blocked block to create this information.”
Amodei, it has been part of anthropic assessments that works in various AI models to evaluate the risk of anthropic regularly in the anthropic regularly. The team looks that the models can create information related to Bioweapons that are not easily found in Google or textbooks. Anthropic positions itself as the AI foundation model provider Safety safety is serious.
Amodei, today Deepseek’s models said that he did not think it was “dangerous” in providing a rare and dangerous information, but he could be in the near future. Although Deepseek praised the team as “talented engineers”, the company recommended the company to “take serious security considerations.”
Amodei also has Supported strong export control In chips in China, it causes China to give a military edge.
Amodei, DeepSEEK model is the test of the DeepSeek model tested and did not provide more technical information about these tests, the Chinatalk interview was not clear. Anthropic did not respond immediately to a request for comments from TechCrunch. What was DeepSeek.
Deepseek’s rise also gave concerns about his safety because he was elsewhere. For example, Cisco Security Researchers He said last week The DeepSEEK R1 could not block any malicious instructions in security tests reaching 100% jailbreak success rate.
Cisco did not mention the Bioweapons, but said that DeepSeek could be obtained to create harmful information about cyber crime and other illegal activities. In addition, Meta Llama-3.1-405B and Openai’s GPT-4O is worth 96% and 86% high failure rates.
It remains to see if the safety concerns have a serious bit of Deepseek fast acceptance. Companies like Boring and Microsoft Connecting R1 to cloud platforms, taking into account the cloud platforms – as ironic enough, Amazon The largest investor of anthropic.
On the other hand, Have a growing list of the countries, Companiesand especially government organizations such as US Navy Pentagon Began to ban DeepSeek.
In the event of this effort or the global growth of Deepseek will continue. In both cases, Amodei says that a new opponent, which is the best EU levels of EU companies.
“A new truth here is a new opponent,” he said. “AI – Anthropic, Openai, Google, Maybe in large companies that can cultivate meta and xa – are now being added to this category.”
