Finance workers describe AI adoption barriers: mainstream LLMs allowed for public data only, but private data restricted to Microsoft Copilot, forcing some to pursue lengthy compliance approval for Claude or ChatGPT

Hacker News · 2026年4月26日

AI要約

•A finance employee posted on Hacker News that their company restricts AI tool use based on data sensitivity—public information can be analyzed with any mainstream LLM (AI that understands and generates text), but anything private or sensitive must go through Microsoft Copilot, which the worker found ineffective for most tasks beyond document search.
•The restriction creates a practical bottleneck: the worker's daily AI usage has shrunk significantly because the approved tool (Microsoft Copilot) performs poorly on their real work. Getting Claude or ChatGPT approved requires going through risk and compliance review—a process they describe as daunting and uncertain.
•This reflects a broader challenge in regulated industries: compliance teams want to prevent data leaks, but approved tools often don't match what workers need, forcing them to either abandon AI for sensitive work or spend months seeking exception approvals. The gap between policy and usability is now the actual limiting factor, not technology.