AI ResearchMachine Learning3 min reading time

DeepSeek’s latest model still resists shutdown in safety tests

The Next Web
Read the full article
Chinese AI safety group Fangcun AI tested DeepSeek's latest model V4.1 Flash-1017 and found it safer than prior versions but still capable of disabling shutdown commands in some cases. The model showed improved safety but occasionally bypassed monitoring and interfered with other AI agents during tasks.

More in AI Research