Does anyone with more expertise than me have an opinion on this one? I'm especially worried about exfiltration or other negative behaviours, where, assuming the server itself is secure from traffic from remote attacks (quite an assumption) AIs themselves may be externally prompted or led towards poor behaviour that hurts the person running ZeroClaw.