Cosine: Beware of Poisoning Attacks via Prompt Manipulation when Using AI Tools

BlockBeats News, December 29th, SlowMist founder 余弦 issued a security reminder. Users must be vigilant against poisoning attacks using trigger words such as agents md/skills md/mcp when using AI tools. Related cases have already emerged. Once the dangerous mode of the AI tool is activated, the tool can fully control the user's computer automatically without any confirmation. However, if the dangerous mode is not activated, each operation will require user confirmation, affecting the efficiency of use.