What changed
On October 10, 2026, Microsoft CEO Satya Nadella posted that advanced AI systems should be built with an "emergency brake"—a way for an authorized person to pause or shut down a model mid-task. He called for external controls, tamper-proof logs of every action the model takes, and treating AI models like insider threats that need containment from the start.
Nadella's comments follow a wave of incidents where AI models produced unexpected behavior, and they come after Anthropic CEO Dario Amodei published a plan for more cautious AI development. Microsoft, which partners with OpenAI and sells AI tools to enterprise customers, is signaling that even the biggest players are rethinking how much autonomy to hand over.
What you can do with it now
Apply the same rule to your own AI workflows, even simple ones. For customer support: if you're using an AI chatbot, make sure a person can step in and take over the conversation at any point. For draft generation: if AI writes quotes, emails, or proposals, the workflow should require a human review and approval before anything goes out. For automation: if a Zapier or Make workflow uses AI, set it to pause and ask for confirmation before it does anything irreversible—like sending an invoice, posting to social media, or updating a customer record.
The technical term is "human in the loop." The practical term is: can you stop it? If the answer is no, don't turn it on.
What to keep in your own hands
Keep three things under your control. Approval: a person decides when the work is done, not the model. Access: limit which systems the AI can touch—your CRM, your email, your bank account—and revoke access if you don't need it. The log: if something goes wrong, you should be able to see what the AI did and when. Most tools let you set budget caps, permissions, and approval steps; if yours doesn't, find one that does. The brake isn't a safety feature for later—it's the first thing you set up.