Application developers and device manufacturers like Google and Apple are building new AI features into their products without offering clear guidance on how to control the information they access. AI chatbots are being rolled out into messaging apps, and AI features added to summarize conversations or suggest text. When the conversation is encrypted, messaging app creators must not decrease the existing security level by allowing automatic leakage through AI features. Users must have an accessible way to disable AI access. 

System-level AI on devices should never be able to access encrypted conversations without opt-in user consent. EFF is pressing companies to develop common sense standards, and publishing guidance to help users understand what’s at stake.