Apriorit CEO Warns Against Post-Launch AI System Failures

Klaudia Zaika, CEO of Apriorit, warns business leaders that treating the release of an AI system as a finish line is a mistake, as critical failures typically emerge after launch. She identifies primary risks including silent degradation and context rot, where models struggle with real-world data, alongside a lack of output quality monitoring for hallucinations and data drift.

Zaika points to legal liabilities resulting from AI-generated errors, citing a 2024 case where Air Canada was ordered to honor a discount hallucinated by its chatbot and a 2026 German court ruling that held Google responsible for false statements in AI Overviews. She also highlights security vulnerabilities such as prompt injection, which OWASP identified as a leading cause of failure in production agentic systems in its 2026 analysis.

To mitigate these risks, Zaika recommends implementing least-privilege access and assigning human owners to critical AI outputs. She emphasizes the necessity of continuous monitoring, a requirement for high-risk systems under the EU AI Act.