AI Visibility and Control recognizes more than 50 AI agents and more than 70,000 skills.
Four classes of exposure the software inventory misses
Bitdefender frames the problem plainly: traditional software inventories miss four specific classes of AI exposure that now matter to defenders. The first is AI services reached through the browser — hosted assistants that show up in user activity but may not be on any approved list. As the product notes, "A service with no findings is not automatically approved; a clean result on a tool your organization never sanctioned is still shadow AI."
Second are coding assistants that run inside editors alongside source repositories, credentials, and configuration files. Third are local model runtimes, MCP servers and agent frameworks that execute on the endpoint itself — often without an installer or any entry in an installed programs list. Fourth are tools configured to reach other systems — databases, code repositories, internal APIs, or third-party model services — where the risk lies in the scope of those connections rather than in visible network traffic.
How AIVC discovers and categorizes what it finds
Discovery happens during the GravityZone risk scan. The GravityZone BEST agent "collects metadata from each endpoint" but, the product specifies, keeps that collection minimized: it records what AI software is present and how it is configured, "not what anyone does with it." AIVC explicitly does not inspect prompts, chat history, model weights, credentials, or full file content.
Found services are assigned to one of 17 functional categories — examples given include LLMs, coding, productivity, Platforms and infrastructure, Agents, Automation, and MCP. This classification is designed to let reviewers filter the inventory by the question they came to answer: a security team concerned with data-handling restricts to LLMs, chatbots, Platforms and infrastructure; a development manager looks at Coding and Tools and skills; a governance review focuses on Agents, Automation, and MCP.
This site is the portfolio.
OSINTSights runs on Cloudflare Workers, D1, R2, and Vectorize, with an AI pipeline on Hetzner ARM. Nubivance designed, built, and operates it. We do the same for clients.
See what we buildTwo console views: AI Services and AI Findings
Within the Bitdefender GravityZone Console, AI Visibility and Control appears as two complementary views. AI Services is the full inventory — every discovered service with its risk score, functional category, related AI assets, open findings count, affected endpoints, state (Active or Ignored), and whether it sits on a watchlist. Opening a service shows mitigation guidance alongside its details.
AI Findings is the prioritized, day-to-day view. Each finding includes a risk score, the related service, affected resources, a link to mitigation guidance, and the endpoint-specific evidence that raised the finding. That evidence is designed to make findings actionable for a SOC analyst: a finding tied to a local model runtime or an MCP server identifies a process executing on a specific host and provides an affected-host list to pivot from in Search and Incident Advisor.
From inventory to enforcement: what AIVC does and what it leaves to existing controls
AIVC establishes presence and configuration risk, but it does not perform enforcement itself. The walkthrough is explicit: the score "orders your queue; it does not decide the outcome." Turning on AIVC "does not block access, remove software, or modify program permissions." Enforcement remains with existing GravityZone controls.
Three enforcement paths are described. Web Content Filtering blocks access to prohibited services once AIVC identifies domains in use. Custom detection rules let administrators block a specific tool wherever it appears by matching process names and attaching actions such as "Kill process." Hardening policies and PHASR handle reducing what an installed tool is permitted to do: PHASR evaluates the agent separately from the user so agent processes carry their own behavioral profile while user restrictions remain tied to the user's session.
Finally, enabling AIVC requires no separate rollout: it runs on the existing Bitdefender Endpoint Security Tools agent and is enabled by editing a current policy. The setting sits under Risk Management in the General section; results appear after the next risk scan.
What this means for security teams, development managers, and governance reviewers
- Security teams and SOC analysts: gain endpoint-specific evidence tying processes and connectors to hosts, letting triage and incident pivoting start from the affected-host list rather than from a domain filter alone.
- Development managers and engineers: see coding assistants and editor extensions surfaced as part of the inventory, including skills that extend agent capabilities — important because skills "arrive from public registries" and can expand an agent’s permissions.
- Governance reviewers and policy owners: receive a categorized inventory and risk scores, but retain the sanctioning decision — a zero score is not an approval, and a high score is a queue signal rather than an automatic removal.
Bitdefender’s AI Visibility and Control positions itself as the discovery and risk layer organizations lack: it finds browser assistants, editor extensions, local runtimes, MCP connectors and agent frameworks; maps them into 17 functional categories; and reports configuration-focused evidence without collecting prompt or content data. The product hands security teams a prioritized list and endpoint evidence; the decision and the enforcement remain, by design, in the hands of administrators and the policies they already run.
