Bitdefender adds AI discovery and risk views to GravityZone

Bitdefender has detailed AI Visibility and Control (AIVC), a discovery and risk-management capability in the GravityZone console for protected Windows endpoints. The feature is designed to identify AI services that conventional software inventories can miss, including browser-based assistants, editor extensions, local model runtimes, Model Context Protocol (MCP) servers and agent frameworks.

AIVC recognises more than 50 AI agents and more than 70,000 skills, the modular packages that extend an agent’s capabilities. It assigns every discovered service to one of 17 functional categories, including LLMs, coding and productivity, and associates findings with affected endpoints and resources.

Discovery beyond installed software

Traditional inventories often depend on installed-program records, yet AI use may leave no such trace. A user can paste information into a hosted chat service in a browser, while a developer can add a coding assistant to an editor. Local runtimes and MCP servers can also be configured without appearing as standard installed applications.

Bitdefender says discovery runs during a risk scan through the GravityZone BEST agent. The agent collects metadata about AI software and its configuration, rather than inspecting prompts, chat histories, model weights, credentials or full file content. For tools linked to databases, code repositories, internal APIs or third-party model services, AIVC records the configured connection scope rather than data moving through that connection.

This distinction is particularly relevant for MCP servers. A connector can be introduced through a small configuration entry and give a model access to external tools. A connection to a read-only reporting view differs from one holding write credentials for a production database, and Bitdefender positions that configuration difference as part of the exposure assessment.

Two console views for inventory and triage

The AI Services view provides the inventory of discovered tools. It shows a service’s risk score, functional category, related AI assets, open findings, affected resources, endpoint locations, Active or Ignored state and watchlist status. Administrators can open an individual service for details and mitigation guidance.

AI Findings is the prioritised risk view. Each finding includes a risk score, the related service, affected resources, mitigation guidance and endpoint-specific evidence. That evidence can show why a finding was raised on one host but not another, while affected-host lists can indicate whether a configuration is isolated or repeated across a team.

Bitdefender stresses that a risk score is not an approval decision. A score of zero means AIVC found no risky configuration, not that the organisation has sanctioned the tool. Teams are therefore expected to review the AI Services inventory against their own approved-tool lists, rather than relying only on findings.

Enforcement remains in existing controls

AIVC does not itself block services, remove software or alter program permissions. Bitdefender points to Web Content Filtering for blocking browser-accessed services, custom detection rules for identifying a named process and applying an automatic Kill process action, and hardening policies and PHASR for restricting the behaviour of installed tools.

PHASR for AI agents evaluates a known agent binary separately from the person using the endpoint, so the agent can receive a behavioural profile distinct from the user session. AIVC can be enabled in an existing policy under General and Risk Management, with results appearing after the next risk scan. For businesses, the practical implication is to use the discovered endpoint inventory to distinguish unsanctioned AI from risky configuration, then apply approval and enforcement decisions through the controls already in place.