AI agent kill switch urged by Okta-led alliance – how businesses could make it work

Okta, Google, Amazon Web Services and Salesforce are among the companies that have formed the Blueprint Alliance, announced this week at Okta's annual Oktane conference. The coalition published a blueprint for agentic visibility, control and governance centered on four questions it says businesses should be able to answer: where are my agents, what can they do, what are they doing, and how do I respond.

The alliance published six operational principles, one of which states that every agent needs an immediate kill switch to suspend or terminate operations, with a clear path to restore function. The group frames the kill switch as a response to problematic agent activity that unfolds at machine speed, shortening the window for defenders. ZDNET cites research from LastPass finding 92% of business admins say AI is already in use across their organization while 27% have an enforced AI governance program, Okta research finding 92% of organizations use autonomous agents and 34% secure them as rigorously as humans, and Gartner finding 13% of organizations believe they have the right AI agent governance in place.

Picus Security associate security research engineer Umut Bayram told ZDNET that organizations cannot respond to attacks unfolding in minutes with processes that take days. The report notes that not all anomalous agent activity is malicious, citing a well-intentioned agent entering an infinite loop that drives excessive LLM billing.

Okta chief product officer Ely Kahn described two scenarios to ZDNET: killing an agent outright to stop behavior before it gets out of control, and putting a new guardrail in place, such as one preventing exfiltration of certain data or changing an agent's permissions. During CEO Todd McKinnon's keynote, attendees saw a demonstration in which a Claude agent asked to forward confidential information from Salesforce to an employee's personal email address was detected by another agent, which deprovisioned the first agent's Salesforce access by revoking its token, notified its human owner and messaged the IT department via Slack. The report says the process completed in seconds.

McKinnon said identity providers cannot necessarily address every aspect of the blueprint and that some non-identity-based telemetry must come from other sources. Okta also showed Shadow AI Agent Discovery for Endpoints and Okta Identity Threat Protection, which aggregates risk intelligence from CrowdStrike, Zscaler, SentinelOne and Palo Alto Networks into a single view. The report ties centralized token revocation to the IETF's Identity Assertion Authorization Grant, credited in large part to Okta director of identity standards Aaron Parecki, with IAAG co-author Brian Campbell of Ping Identity.