Okta Targets AI Agent Identity with Kill Switches

Okta CEO Todd McKinnon is betting the company's future on managing the identity of AI agents, a strategic move designed to counter the threat of customers building their own software tools. In a recent interview, McKinnon discussed the emerging concept of "agent identity" and the necessity of security measures like a "kill switch" to manage these autonomous systems within corporate environments [1].

Key takeaways

• McKinnon describes the "SaaSpocalypse" as a threat where customers use AI to build tools instead of paying SaaS fees [1].
• Okta is exploring "agent identity," treating AI agents as entities between people and systems that require management [1].
• The CEO suggests installing a "kill switch" at the agent level to maintain security as platforms like OpenClaw rise [1].
• McKinnon views the shift toward AI as a disruption larger than cloud computing, despite being "paranoid" about the risks [1].

Facing the 'SaaSpocalypse' Threat

McKinnon, who leads a platform with a $14 billion market cap and $3 billion in revenue, expressed concern about the "SaaSpocalypse" on a recent earnings call [1]. He described this phenomenon as a scenario where companies utilize large language models to "vibe-code" their own internal tools rather than paying subscription fees to established SaaS providers [1]. Despite Okta's established base of 20,000 customers, McKinnon admitted to being "paranoid" about this trend, characterizing the rise of AI and agents as a disruption potentially larger than cloud computing [1]. He believes the total addressable market for software is expanding, but capturing that value requires the company to adapt to significant technological changes [1].

Securing Hybrid Teams of Humans and Agents

To address these challenges, Okta is focusing on "agent identity," a concept McKinnon describes as existing somewhere between a person and a system [1]. This approach aims to secure corporate environments as employees adopt platforms like OpenClaw and potentially hand over credentials to hardware like Mac Minis to perform tasks [1]. McKinnon suggested that simply installing a "kill switch" at the agent level might be necessary to ensure users, platforms, and data remain safe [1]. The shift also raises complex organizational questions, as companies may soon need to manage hybrid teams comprising both humans and AI agents [1].

Why it matters

The push toward agent identity highlights a critical pivot for enterprise security as AI moves from theoretical to operational use. McKinnon believes that while the "pie" for technology is expanding greatly, success depends on managing the security credentials of non-human actors [1]. This evolution suggests that identity management will become a central component of how enterprises integrate and control AI workflows, ensuring that autonomous agents can be effectively managed or terminated if necessary [1].