Amazon staff identified significant budget overruns on artificial intelligence projects.
One failed deployment using Anthropic’s Claude AI model to match author details cost $1.8 million. This specific project exceeded its budget by 860%. The error remained undetected for five months.
Engineers reported additional unexpected costs exceeding $500,000 on separate projects. Another project accumulated $134,000 in unplanned expenses.
Simple coding mistakes drove these expenses when applied to AI models. Amazon is now developing automated guardrails to prevent future overruns. The company aims to improve cost efficiencies across new technology deployments.