[P0] Terms §5 overclaims: "cannot read network traffic" is false for first-party inference #156
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
PRE-LAUNCH. §5 'The Network Layer' says Neuron LLC 'cannot read, inspect, retain, or produce the contents of network traffic as an architectural constraint.' That is FALSE for the first-party Neuron Inference path: you cannot run inference on ciphertext — the server terminates TLS and the model reads the plaintext prompt. It holds only for pass-through/relay to third-party providers. Internal contradiction: §4 correctly uses POLICY language ('will not store/train/share') while §5 uses CAPABILITY language. FIX: narrow §5 to 'encrypted in transit; for relayed third-party traffic we do not retain/inspect; for Neuron inference the request is processed in memory and never stored/logged/trained-on/shared.' Restore 'cannot' only when confidential-compute/TEE inference ships. ALSO add a feedback/telemetry clause (none today): response thumbs stored locally and never transmitted; the feedback door sends only typed text, user-initiated. Spec: docs repo (PR #1).