In this paper we propose an architecture which combines watermarking with traditional cryptographic key agreement protocols to establish an authenticated or encrypted channel in teleconferencing systems. Technically the proposed method embeds messages of the key agreement protocol within an audio or video stream and is based on the assumption that the human communication partners can recognize each other easily; the watermark establishes a close coupling between the cryptographic key exchange messages and the media stream. We argue that the security of the scheme is based on a yet unexplored security property of digital watermarks; furthermore we present preliminary research results that suggest that this property holds in standard watermarking schemes.