AGENTFORCE-12.1: Latent Memory Poisoning in Prompt Template

🚨 Critical · Einstein Copilot Studio Configuration

Detects GenAiPromptTemplates that use Conversation Memory ({!Conversation.*}) and ground from Apex data providers that lack input sanitisation. Unsanitised data stored in conversation memory can activate as a latent poisoning trigger when retrieved in a future conversational context (ref: SSRN-6372438 'Latent Memory Poisoning').

Details

Field	Value
Rule ID	AGENTFORCE-12.1
Severity	Critical
Category	Einstein Copilot Studio Configuration
Compliance	EU_AI_ACT_HIGH_RISK, NIST_AI_RMF

Remediation

Refer to the SquireX documentation for remediation guidance specific to AGENTFORCE-12.1.

See Also

• Rule catalog index
• SquireX scan documentation