After SalesBleed: three practical controls for Agentforce prompt injection protection
SalesBleed之后的Agentforce提示注入防护
AI summary
The author discusses Agentforce's prompt injection defenses against external CRM data, referencing the SalesBleed research disclosed by Zenity Labs. The article emphasizes distinguishing untrusted records from operational instructions, reviewing agent permissions, and continuously checking defense configurations. It states that the disclosed issue has been fixed and reminds that existing records and subsequent versions still need evaluation; these recommendations are part of the author's defensive practice notes.
Why it was selected
The author provides three actionable Agentforce hardening measures, along with trigger and fixed query action code, suitable for teams running public forms plus Agent to check against.