Now Assist Guardian - Prompt Injection - How Does it Work?

[JustinD56844010]

We are enabling Now Assist in Virtual Agent.

We are also enabling Now Assist Guardian - Prompt Injection.

We are struggling with understanding what ServiceNow considers a malicious prompt to be?
If a user using the Virtual Agent in Employee Center asks the Virtual Agent for another employee's home address - is that considered a malicious prompt?

The table that stores the logging for the malcious prompts, sn_nowassist_admin_sys_gen_ai_export_data, is not easily readable so it's hard to figure out how 'Prompt Injection' works.

[Tanushree Maiti]

Hi @JustinD56844010 

Check these resources if helps.

Configuring Now Assist Guardian: Offensive Content & Prompt Injection Protection

Protecting Your AI with Now Assist Guardian: Offensive Content Stops Here

ServiceNow Now Assist Guardian: Enterprise Protection That Accelerates Deployment

Now Assist Guardian FAQ

NowAssist Guardian: A Comprehensive Guide for Safer AI Use in ServiceNow

https://www.youtube.com/watch?v=AoEjEvDR088

https://www.youtube.com/watch?v=OFt1XcJ_P18

https://www.youtube.com/watch?v=-MQDRe-o24o

https://www.youtube.com/watch?v=jKBSN26muQI

[hugohuyghue]

Hi @JustinD56844010
I do think the resources Tanushree Maiti suggested are very helpful and will be able to assist you.