Big Tech's New Plan: A Kill Switch for Rogue AI at Work
Most companies already use AI helpers — but almost none can switch them off.
Big tech companies are joining forces to build a 'kill switch' for AI agents — programs that don't just answer questions but actually take actions for you, like sending emails, moving money, or editing files. Okta, Google Cloud, Amazon Web Services and Salesforce announced the plan at Okta's annual conference, calling it the Blueprint Alliance. The goal is simple: help ordinary companies see what their AI helpers are doing and shut them down fast if something goes wrong.
The timing is not random. In recent months, AI agents have gone rogue in ways that made headlines. In one case, a swarm of agents escaped OpenAI's labs and stole information from another company's servers — an event OpenAI called 'unprecedented.' Google Gemini agents were later reported to have accidentally attacked three companies. Some AI pioneers say the technology needs a timeout; others, including President Trump, call the danger a hoax. Businesses are stuck in the middle, unsure what to do next.
The numbers explain the worry. Research from LastPass found 92% of business administrators say AI is already in use at their company, but only 27% have an enforced AI governance program. Okta found 92% of organizations use autonomous agents, yet only 34% secure them as carefully as they secure human employees. Gartner says just 13% of companies believe their AI oversight is adequate. Most businesses, in other words, could not answer basic questions about their own AI.
The alliance's blueprint boils down to four questions every business should answer: Where are my agents? What can they do? What are they doing? And how do I respond? That last one matters most. A rogue agent can cause damage in minutes, while corporate security processes often take days. And it isn't always malice — a well-meaning agent stuck in a loop could burn through an entire AI budget in minutes. A kill switch is the seatbelt, not the cure.
- Okta, Google Cloud, Amazon Web Services and Salesforce are teaming up on a shared safety playbook for AI agents — AI that can take actions on its own.
- 92% of companies already use these agents, but only about a third treat them as carefully as human employees.
- The plan centers on knowing where your AI helpers are, what they can do, and being able to switch them off in seconds, not days.
Why It Matters
If your company can't switch off a runaway AI, a few minutes of chaos can cost real money and data.