This research evaluates how large language models handle deprecated APIs when generating code suggestions, revealing a significant challenge for AI-powered development tools.

• LLMs trained on large code repositories frequently suggest outdated and deprecated APIs
• The paper presents a systematic evaluation framework to measure this problem
• Results show LLMs need better mechanisms to stay current with rapidly evolving libraries
• The findings highlight engineering risks when relying on AI coding assistants without validation

For engineering teams, this research underscores the need for caution when adopting LLM-based code completion tools, as they may introduce technical debt or security vulnerabilities through outdated APIs.

LLMs Meet Library Evolution: Evaluating Deprecated API Usage in LLM-based Code Completion