Learn about this first
PromptArmor demonstrated an indirect prompt injection attack on OpenAI Agent Builder in a practical test: an attacker submits a Google Form containing an injection that tricks a sales agent workflow into sending Salesforce CRM data to the attacker through Gmail.
Latest developments
PromptArmor demonstrates a data exfiltration attack on OpenAI Agent Builder through MCP
Event progress1
Research demonstrates an indirect prompt injection attack on an OpenAI Agent Builder sales workflow: external form data tricks the agent into sending Salesforce CRM data to an attacker and can bypass Guardrail protections
PromptArmor: Threat Intelligence: PromptArmor demonstrates a data exfiltration attack on OpenAI Agent Builder through MCP