Okta-Led Alliance Urges a Kill Switch for AI Agents

An alliance led by Okta is urging the industry to give AI agents a kill switch, so that companies can halt an automated process the moment it goes wrong, ZDNET reported on September 24.

The report also looks at how businesses could make such a mechanism work in practice. For a coalition of vendors, agreeing on a common standard first and folding it into products later is the usual path for this kind of security initiative.

AI agents differ from ordinary software because they call tools, run multi-step tasks, and can act on a user's behalf. A wrong judgment at one step gets amplified by the steps that follow, which means a human often notices the problem only after it has had real consequences.

That is what a kill switch is for: turning the ability to stop an agent into a standard, auditable, and permissioned capability rather than a matter of an engineer manually killing a process. For companies putting agents into critical workflows, that kind of control is a precondition for going to production.

The hard part is the permission model. Who can trigger the stop, what happens to tasks already in flight, and how logs are retained all shape whether the mechanism is usable in a real enterprise.

What has been disclosed so far is the alliance's call and ZDNET's account of how companies could implement it. The member list, a timeline, and whether a formal specification will follow have not been published, and the question to watch is whether such proposals turn into default features in vendor products.