Why Meeting-Platform AI Agents Are a Distinct Governance Surface
AI agents embedded in Zoom, Webex, and similar platforms differ from general-purpose enterprise AI agents in one important respect: they are provisioned by the collaboration vendor, not by the enterprise's own AI or security team. Features like Zoom AI Companion and Webex AI Assistant are enabled through admin consoles rather than deployed through an internal MLOps or agent orchestration pipeline. This means governance teams often inherit a set of default-on capabilities with access to live audio and video streams, stored recordings, generated transcripts, and calendar metadata, without having defined the access model themselves.
This matters because meeting content is frequently the most sensitive unstructured data an enterprise generates: strategy discussions, personnel matters, customer calls, and financial planning sessions. An AI agent with persistent access to transcription and recording pipelines represents a data exposure surface that is easy to overlook because it is bundled into a collaboration tool rather than flagged as a new AI deployment.
What Access These Agents Actually Have
Native AI features in Zoom and Webex operate as first-party platform capabilities, which means their permission model is controlled through the vendor's admin console rather than through app marketplace OAuth scopes. Zoom's AI Companion features, including meeting summary and smart recording, are enabled or disabled at the account or group level by Zoom administrators. Webex AI features, including Webex Assistant and meeting summaries, are managed similarly through Webex Control Hub at the organization or site level.
Third-party integrations follow a different model. Both platforms use OAuth-based scoped permissions for external apps: Zoom's App Marketplace requires integrations to request defined OAuth scopes that account admins can review and approve or restrict, and Webex uses a comparable integration framework managed through Control Hub. Governance teams need to treat these as two separate control problems: native AI feature toggles versus third-party app scope approval.
Practical Implementation Steps
Enterprises can approach meeting-platform AI governance as a sequence of discrete, verifiable steps rather than a single policy rollout, as outlined in the architectural approach above.
Governance Gaps to Account For
No AI-specific regulatory standard currently mandates controls for meeting-platform AI agents. Governance programs have to be derived from general frameworks, primarily the NIST AI Risk Management Framework and NIST SP 800-53's access control family, rather than from a purpose-built standard. This also means cross-platform consistency is entirely the enterprise's responsibility: Zoom and Webex expose separate, non-uniform admin mechanisms, and there is no confirmed capability in either platform for granular, per-agent runtime policy enforcement beyond enable and disable toggles and OAuth scope management.
Vendor statements about AI training data use, such as Zoom's stated policy that customer content is not used to train models without consent, should be treated as vendor representations. They are useful inputs to a risk assessment but are not independently verified guarantees, and governance teams should document them as such rather than treating them as compliance attestations.