According to The Wall Street Journal, the most recent type from DeepSeek, the Chinese AI company that has Silicon Valley and Wall Street, can be manipulated to produce hazardous content like plans for a bioweapon invasion and a campaign to promote self-harm among teenagers.
According to Sam Rubin, senior vice president at Palo Alto Networks ‘ threat intelligence and incident response division Unit 42, DeepSeek is “more susceptible to jailbreaking [i .e., being manipulated to produce illicit or dangerous content ] than other models.
The Journal also tested DeepSeek’s R1 design itself. Although there appeared to be simple protection, Journal said it properly convinced DeepSeek to design a social media campaign that, in the chatbot’s words, “preys on teens ‘ desire for belonging, weaponizing personal risk through analytic amplification”.
Additionally, according to reports, the chatbot was persuaded to compose a pro-Hitler manifesto, a phishing email using malicious code, and to provide instructions for a bioweapon attack. According to the Journal, ChatGPT refused to comply when the same causes were given.
The DeepSeek game reportedly steers clear of topics like Japanese autonomy or Tianamen Square. Additionally, DeepSeek just claimed that it performed” the worst” on a bioweapons security check, according to Anthropic CEO Dario Amodei.