In point-to-point 1-on-1 WebRTC calls, DTLS-SRTP encrypts the media directly between the two endpoints. However, in group video conferences with 3 or more participants, point-to-point mesh requires O(N^2) bandwidth and CPU encoding, which quickly overwhelms mobile devices. Group calls rely on Selective Forwarding Units (SFUs) to route streams. Because DTLS-SRTP terminates at the media server, the SFU possesses the decryption keys and can inspect, record, or modify all unencrypted audio and video payloads. True E2EE requires frame-level encryption before RTP packetization, ensuring the SFU acts as a blind relay.

SFrame (RFC 9605) is a lightweight, frame-level encryption format designed specifically for real-time media. Instead of encrypting at the RTP packet layer, SFrame encrypts each complete encoded video or audio frame (using AES-GCM or AES-CTR with HMAC) immediately after the codec outputs it. SFrame prepends a tiny header (containing a Key ID and frame counter) and appends an authentication tag (typically 80 or 128 bits). The encrypted frame is then passed to the browser's standard WebRTC packetizer. The SFU receives standard RTP packets, inspects unencrypted RTP sequence numbers and timestamps to route packets to subscribers, but has zero ability to decrypt the underlying video/audio data.

WebRTC Insertable Streams (via the RTCRtpScriptTransform API) expose encoded audio and video chunks as standard WHATWG Streams inside a dedicated Web Worker thread. On the sending peer, an TransformStream intercepts each RTCEncodedVideoFrame before packetization, reads frame.data into an ArrayBuffer, applies SFrame encryption, and writes the ciphertext back. On the receiving peer, another worker stream intercepts the incoming encoded frames, verifies the SFrame authentication tag, decrypts the payload, and passes the plaintext frame to the browser's video decoder.

SFrame was engineered for minimal bandwidth impact. A typical SFrame header occupies only 3 to 5 bytes (using variable-length integer encoding for Key ID and Frame Counter), and the authentication tag adds 8 to 16 bytes. Total overhead is only ~11 to 21 bytes per frame. At 30 frames per second, SFrame adds less than 0.5 KB/s of network overhead—virtually undetectable. Using hardware-accelerated WebCrypto AES-GCM, frame encryption takes less than 0.2 milliseconds, adding imperceptible latency to real-time communication.