Claude and ChatGPT Connectors Change Every 9 Minutes (www.promptarmor.com)

Recent tracking of connectors for AI tools like Claude and ChatGPT reveals that their capabilities are changing approximately every nine minutes. During a six-week monitoring period, 37% of connectors underwent modifications, including expanded permissions and new tool integrations that significantly alter how these AI systems can interact with third-party applications. Notably, users have requested connections to external apps, broadening the functionality of these AI agents but also raising concerns about security and data governance. For example, connectors have transitioned from read-only to write-capable, allowing them to perform actions such as deleting files or sending emails, thus introducing new risk factors for organizations. These frequent changes highlight a pressing need for organizations to continuously assess the risks associated with AI connectors, especially given the potential for indirect prompt injections and untrusted data processing. As connectors modify their permissions and tool behaviors without a formal notification process, organizations are urged to develop governance strategies that can keep pace with this rapid evolution. Failure to adapt could lead to unauthorized access to sensitive data or unintended actions taken by AI agents, underscoring the critical importance of vigilant connector management in the AI/ML landscape.