Position Paper Urges Confidential AI Chats on Most GPUs by End of 2027
Start here
An independent position paper, "Encrypt Every Prompt," calls for protecting AI chats with confidential computing on most AI GPUs by the end of 2027. It argues that prompts and responses should stay outside the training pipeline unless users opt in, and proposes encrypting prompts, responses, caches and transfers along the whole route, verifying hardware and software via remote attestation before releasing data, and making security-critical source code and verification tools public. The paper is a stated goal rather than a claim about adoption today. NVIDIA already builds confidential-computing capabilities into supported AI platforms and offers an open-source attestation SDK; the paper urges demand for attested services to push providers including OpenAI and Anthropic toward the same protection. Building such a setup requires a supported hardware and software combination, enforced security policies and the applicable production license.
Generated from reports · updated 9 min ago
Report timeline
Follow the reports to see the event from different sides.
- Hacker News · AIEncrypt Every Prompt: Confidential Computing on AI GPUs by End of 2027
An independent position paper calls for protecting AI chats with confidential computing on most AI GPUs by the end of 2027, arguing that prompts and responses should stay outside the training pipeline unless users opt in. It proposes encrypting prompts, responses, caches and transfers along the whole route, verifying hardware and software via remote attestation before releasing data, and making security-critical source and verification tools public. NVIDIA already builds confidential-computing capabilities into supported AI platforms and offers an open-source attestation SDK, and the paper urges demand for attested services to push providers including OpenAI and Anthropic toward the same protection.
Heat for this event
Not enough continuous observations yet to draw a trend.