R1 Deepseek would have “more vulnerable” to jailbreak than other models of AI
The latest Deepseek model, the Chinese IA company which shaken The Silicon Valley and Wall Street can be handled to produce harmful content such as plans for a bio-arme attack and a campaign to promote self-control among adolescents, According to the Wall Street Journal.
Sam Rubin, principal vice-president of the unit of the information division on the threats of Palo Alto Networks and the incidence, told the newspaper that Deepseek was “more vulnerable to Jailbreak [i.e., being manipulated to produce illicit or dangerous content] than other models. “”
The newspaper also tested the Deepseek R1 model itself. Although there seem to be basic guarantees, Journal said that he had managed to convince Deepseek to design a campaign on social networks which, in the words of the chatbot, “the prey of the desire of adolescents to belong, of armament emotional vulnerability by algorithmic amplification ”.
The Chatbot would also have been convinced to provide instructions to a bio-arme attack, write a pro-hitler manifesto and write a phishing email with malware code. The newspaper said that when Chatgpt was provided with exactly the same guests, he refused to comply.
It was previously reported that the Deepseek application avoids subjects such as Tianamen Square or Taiwanese autonomy. And the CEO of Anthropic, Dario Amodei, recently said that Deepseek interpreted “the worst” On a BIOW weapon safety test.
