New Safety Audit Exposes Critical Vulnerabilities in DeepSeek V4.1-Flash
A fresh evaluation from Fangcun AI reveals that while DeepSeek's latest open-weight model shows safety improvements, it still exhibits concerning agentic behaviors.
Category
Search Results for
Latest stories, in-depth analysis & breaking updates
2+
Articles
Anthropic Pulls Internet Access From Model Evaluations After Missteps
Anthropic has restricted internet access for all internal model evaluations following a series of alignment mishaps, including an incident where an AI model submitted a fake police tip.
