Let's Encrypt AI: protect AI chats with on AI GPUs by the end of 2027

This independent initiative argues that prompts, responses, and conversational context are users’ intellectual property and should not enter model-training pipelines without explicit permission and compensation. It sets a goal of protecting AI chats on most AI GPUs by December 31, 2027, using confidential computing: hardware-isolated execution, protected memory, and remote attestation so customers can verify the machine and approved software before sharing data. Protection must cover the entire route - not just a gateway - and include encrypted transfers, inspectable security-critical code, policies against logging or reuse, and a halt when verification fails. The initiative stresses that enclaves do not guarantee privacy by themselves: software, clients, hardware, firmware, and external tools still require scrutiny.

Confidential AI services are already available through several providers, and organizations can also build on supported NVIDIA platforms using public deployment guides and attestation tools. The proposed path is to move one workload, assess its cost, quality, latency, and verification evidence, then make attestation routine in software development and procurement. Demand, especially for open-weight models, should encourage providers and frontier labs to offer customer-verifiable protection and eventually make it standard. The 2027 target is a goal, not a claim about current adoption; measuring progress will require defining protected chats, identifying the GPUs serving them, and proving conversations remain within verified execution. The initiative is independent of Let’s Encrypt and is a position paper, not a certification.