Contain an AI agent for security dashboard in AI control tower in servicenow

navyamr1
Mega Contributor

Hi Team, I am working on Security dashboard in Super Intelligence Control Tower. I want to contain an AI agent manually. I am unable to find options "Deactivate AI agent" in actions for any AI assets in lifecycle. I have completed the lifecycle onboarding, change requests and offboarding lifecycle. But still I was unable to find how to deactivate and reinstate an AI agent. Can anyone please help here.

Also how to create Privileged AI assets and is it possible to manually "create a Security event" for test purpose. 

1 ACCEPTED SOLUTION

sanjaygupta
Tera Guru

Hello @navyamr1 
You won’t find “Deactivate AI agent” in the lifecycle actions, because containment isn’t part of the onboarding, change request or offboarding lifecycle. It’s a separate feature called kill switch protocol, and it lives on the Security page. It only shows up when a few things are in place.

The role you need is AI steward (sn_ai_governance_ai_steward). Newer versions of AI Control Tower also add an AI Security Operator role, which lets someone deactivate and reinstate agents without full AI steward access. You also need a connection between ServiceNow and each platform that hosts your agents, such as AWS Bedrock, Bedrock Agent Core, Azure AI Foundry or Gemini Enterprise. Okta is optional. Without those connections the option won’t appear. Here is how you can check and set it up. Manage AI agents using kill switch protocol +2

  • Make sure your user has the AI steward or AI Security Operator role.
  • Go to the Security page in AI Control Tower and open the security connections setup.
  • Add a connection for the platform your agents run on, and add Okta if you use it.
  • Open the Control Enforcement Points tab and refresh the page. servicenow

Once that’s done, you contain an agent from a security event, not from the agent’s own record. The documented path is Govern > Security > Overview > Your top recommendations, or Activity Center > Recommendations, then opening the Critical AI asset events item. servicenow

  • Open the critical event for the agent.
  • Review the pane showing the detected activity and the suggested next action.
  • Select Deactivate, enter a reason, and confirm.

To reinstate, resolve the critical security task for that agent first.

  • Go to Security > Overview > Contained AI agents.
  • On the agent’s row, open More actions and select Reinstate. servicenow
  • Enter a reason and confirm.

Your screenshot explains what you’re seeing now. Your only recommendation is “Low AI Asset events”, which has zero events, so there’s nothing to act on. Contained AI agents also shows 0 for the same reason.

For privileged AI assets, I couldn’t find any documented way to create one manually. As far as I can tell, privileged status is detected, not set. The Privileged AI agents tile is described as agents with elevated access permissions, and your agent map is built from identity data.

  • Connect the platform or identity provider where the agent runs.
  • Give the agent’s identity elevated permissions in that platform, for example an admin-level role.
  • Let the sync run, then check the Privileged AI agents tile again.

I’m not certain of the exact sync timing, so treat that as a best guess.

For a test security event, I also couldn’t find a “create security event” button or a documented manual option. Events come from runtime and post-runtime security analysis of agent activity, so the realistic test is to trigger one. In version 1.9.5 and later, the sampling rate and max skill calls settings control how much activity is analysed. servicenow

  • In a sub-production instance, run a test agent that is connected and onboarded.
  • Send it inputs that should trip your security checks, such as prompt-injection-style text.
  • Check that the sampling rate allows the activity to be analysed.
  • Watch the Post runtime and Runtime tabs for the event, then open it from Recommendations.

Test in sub-prod only, since a real containment revokes the agent’s credentials on the connected platform and can disrupt a live agent.

If you tell me your AI Control Tower version, which platforms your agents run on, and whether you’re on the new or legacy UI, I can narrow this down. Some of the newer features are only in the redesigned interface.

Sources:

View solution in original post

2 REPLIES 2

sanjaygupta
Tera Guru

Hello @navyamr1 
You won’t find “Deactivate AI agent” in the lifecycle actions, because containment isn’t part of the onboarding, change request or offboarding lifecycle. It’s a separate feature called kill switch protocol, and it lives on the Security page. It only shows up when a few things are in place.

The role you need is AI steward (sn_ai_governance_ai_steward). Newer versions of AI Control Tower also add an AI Security Operator role, which lets someone deactivate and reinstate agents without full AI steward access. You also need a connection between ServiceNow and each platform that hosts your agents, such as AWS Bedrock, Bedrock Agent Core, Azure AI Foundry or Gemini Enterprise. Okta is optional. Without those connections the option won’t appear. Here is how you can check and set it up. Manage AI agents using kill switch protocol +2

  • Make sure your user has the AI steward or AI Security Operator role.
  • Go to the Security page in AI Control Tower and open the security connections setup.
  • Add a connection for the platform your agents run on, and add Okta if you use it.
  • Open the Control Enforcement Points tab and refresh the page. servicenow

Once that’s done, you contain an agent from a security event, not from the agent’s own record. The documented path is Govern > Security > Overview > Your top recommendations, or Activity Center > Recommendations, then opening the Critical AI asset events item. servicenow

  • Open the critical event for the agent.
  • Review the pane showing the detected activity and the suggested next action.
  • Select Deactivate, enter a reason, and confirm.

To reinstate, resolve the critical security task for that agent first.

  • Go to Security > Overview > Contained AI agents.
  • On the agent’s row, open More actions and select Reinstate. servicenow
  • Enter a reason and confirm.

Your screenshot explains what you’re seeing now. Your only recommendation is “Low AI Asset events”, which has zero events, so there’s nothing to act on. Contained AI agents also shows 0 for the same reason.

For privileged AI assets, I couldn’t find any documented way to create one manually. As far as I can tell, privileged status is detected, not set. The Privileged AI agents tile is described as agents with elevated access permissions, and your agent map is built from identity data.

  • Connect the platform or identity provider where the agent runs.
  • Give the agent’s identity elevated permissions in that platform, for example an admin-level role.
  • Let the sync run, then check the Privileged AI agents tile again.

I’m not certain of the exact sync timing, so treat that as a best guess.

For a test security event, I also couldn’t find a “create security event” button or a documented manual option. Events come from runtime and post-runtime security analysis of agent activity, so the realistic test is to trigger one. In version 1.9.5 and later, the sampling rate and max skill calls settings control how much activity is analysed. servicenow

  • In a sub-production instance, run a test agent that is connected and onboarded.
  • Send it inputs that should trip your security checks, such as prompt-injection-style text.
  • Check that the sampling rate allows the activity to be analysed.
  • Watch the Post runtime and Runtime tabs for the event, then open it from Recommendations.

Test in sub-prod only, since a real containment revokes the agent’s credentials on the connected platform and can disrupt a live agent.

If you tell me your AI Control Tower version, which platforms your agents run on, and whether you’re on the new or legacy UI, I can narrow this down. Some of the newer features are only in the redesigned interface.

Sources:

Hello @sanjaygupta , thank you for the response. It was very helpful.