ServiceNow calls for a measured response to rogue AI agents

As AI agents become more prevalent within enterprises, containment emerges as a crucial governance concern, according to ServiceNow Inc. The company is positioning itself as the control tower for enterprise AI, which includes managing how organizations govern deployed agents. Bhakti Pitre, vice president of AI platform security product at ServiceNow, explained that the "kill switch" for a rogue agent is not simply a binary on/off switch.

Pitre, speaking at Okta's Oktane event, emphasized that the decision to pause or revoke an agent's access should consider both risk and the business processes it supports. ServiceNow outlines AI governance in five steps: discover, observe, govern, secure and measure. Containment measures should be proportional to the level of risk.

For instance, an agent that stops generating results might only require a pause and investigation, whereas an agent authorized to offer discounts could be completely shut down if it is prompted to offer 100% discounts after being subject to prompt injection. ServiceNow can link an agent to the relevant business processes and utilize Veza's identity security technology to adjust excessive permissions.

The company is collaborating with Okta to secure agent session tokens and identities. Pitre concluded by stressing the importance of industry collaboration to democratize AI for widespread use, highlighting the need for all vendors to contribute to the solution.