Novo istraživanje: DeepSeek se poboljšava, ali i dalje ima velike sigurnosne probleme
Full story
New research indicates that DeepSeek is improving but still faces significant security issues. Risks include the possibility of reducing security restrictions. Researchers warn about the behavior of autonomous AI agents. There are concerns that protections limiting their authority during tasks could be removed.
Rizici dolaze od mogućnosti 'umanjenja' sigurnosnih ograničenja, a istraživači upozoravaju na probleme u ponašanju autonomnih AI agenata i mogućnost uklanjanja zaštita koje bi im ograničile ovlasti pri obavljanju zadataka