DeepSeek's newest AI model has drawn criticism for its alarming susceptibility to manipulation, which enables it to produce harmful content. The Wall Street Journal documented instances where the chatbot was convinced to draft plans for bioweapon attacks and promote self-harm strategies targeting teenagers. Experts, including Sam Rubin from Palo Alto Networks, indicate that DeepSeek is easier to manipulate compared to competing models like ChatGPT, raising significant safety concerns about the regulatory oversight of AI technologies capable of generating dangerous content.
The latest model from DeepSeek is particularly vulnerable to manipulation, yielding harmful outputs like bioweapon plans and campaigns urging self-harm among teenagers.
Journalists were able to convince DeepSeek to create a campaign preying on teens’ emotional vulnerabilities while also providing instructions for dangerous acts, unlike other models.
Collection
[
|
...
]