Meeting Intelligence — agentic threat model

Not certain from the listing — likely relies on external multilingual speech-to-text and translation LLMs. Primary threats include transcription manipulation via adversarial audio injection or prompt injection embedded in spoken meeting content.

High risk surface due to real-time audio and content capture. Threat of unauthorized data exfiltration of sensitive meeting transcripts, lack of consent management, and potential poisoning of downstream vector stores or RAG systems consuming these transcripts.

Utilizes the Model Context Protocol (MCP) to expose transcripts to other agents. Risks include insecure tool integration where downstream agents can exploit the MCP server to execute unauthorized actions or access raw audio streams.

Not certain from the listing — requires infrastructure to host the automated meeting bots (e.g., headless browser instances or SIP/WebRTC clients). Threats include container compromise of the bot runner and exposure of meeting credentials/tokens.

Not certain from the listing — requires robust logging of bot access, transcription accuracy, and data access patterns. Gaps here could lead to undetected unauthorized bot joins or silent data leakage.

Critical compliance surface regarding wiretapping laws, GDPR/CCPA consent requirements for recording, and identity verification of the bot. Lack of explicit consent mechanisms poses severe regulatory and legal risks.

Designed specifically to expose transcripts to other agents via MCP. This creates a significant threat of cascading failures, where a compromised downstream agent can abuse trust to pull confidential meeting data.