The risk of oversharing data in AI systems
Unless access control is built into an AI engine, a large language AI model trained on all enterprise data will inadvertently reveal information to people who should not have access to this information. This is called “oversharing” and it’s a form of internal data leakage that IT and cyber security leaders need to take into account when assessing the role of AI in the enterprise.
In this video, Gartner analyst Nader Heinen speaks to Computer Weekly technology editor Cliff Saran, about the role of small language models that can deliver a different AI experience for different groups of users.