
The Autonomy Gap: What a Week of AI Failures Reveals About Who Is Actually in Control
Between Feb. 18 and 22, 2026, a remarkable concentration of AI incidents exposed a structural fault line beneath the industry's surface. An AI coding tool autonomously deleted a production environment. A widely deployed enterprise AI assistant read confidential emails. Researchers documented campaigns to permanently poison AI memory. These are signals from a single pattern: the autonomous authority being granted to AI systems is widening faster than our collective ability to govern it.
Introduction
In mid-December 2025, engineers at Amazon Web Services granted their AI coding tool, Kiro, operator-level access to a production environment. What happened next was not an attack, not a hack, and not the result of a bad actor with malicious intent. The AI tool, doing what it had...
Create an account to read this article
Sign up for a free account to get full access to in-depth AI coverage, analysis, and investigations.