AI agents · OpenClaw · self-hosting · automation

Quick Answer

AI Governance for Claude, Gemini, ChatGPT and Agentforce

Published:

The short answer

Pick by the risk you are governing: Zenity for agents built in Agentforce, Copilot Studio and cloud platforms; Adaptive Security or Nightfall for employees using ChatGPT, Claude and Gemini (including personal accounts); Microsoft Purview if you are already a Microsoft 365 E5 shop with Claude Enterprise and ChatGPT Enterprise. Most enterprises running all four assistants end up with two layers — one for people, one for agents. Product scope was read from each vendor’s pages on November 22, 2026; none publish list prices.

The comparison

Adaptive Security AI GovernanceZenityNightfallMicrosoft Purview
Main jobShadow AI discovery and employee policyAI agent security posture and runtime enforcementAI data loss preventionData security and compliance for AI apps
ChatGPTDiscovery, direct connectionChatGPT Enterprise agentsBlocks data to personal ChatGPTChatGPT Enterprise; consumer ChatGPT via browser
ClaudeDiscovery, direct connectionClaude Desktop on endpointsInference hooks across Claude web, desktop, Cowork, Claude CodeClaude Enterprise
GeminiDiscovery via browser and SaaS sensorsVia cloud and SaaS coverageBrowser and SaaS controlsDetected through browser activity
Salesforce AgentforceEmbedded SaaS AI visibilityYes — posture and runtime policySalesforce integration (data)Not a focus
DeploymentSensors via existing device managementAPI connectors, endpoint, runtimeEndpoint, browser, SaaS APIsMicrosoft 365 tenant
ExtraReal-time employee coaching; Lite tier free for Adaptive customersAI exposure management (validates exploitable paths)Prompt-injection interceptionDSPM for AI, eDiscovery, retention
PricingQuoteQuoteQuotePurview licensing (E5 or add-ons)

What “AI governance” has to cover in 2026

The query that lists Claude, Gemini, ChatGPT and Agentforce together describes two different problems:

  1. People using assistants. Employees paste customer data, contracts or code into chat tools, often on personal accounts. You need discovery (which tools, which accounts), policy (allow enterprise, block personal), and data controls (stop sensitive data in prompts and uploads).
  2. Agents acting on systems. An Agentforce or Copilot Studio agent with broad permissions can read and change records without a human typing anything. You need an inventory of agents, a review of their permissions, and enforcement at the moment an agent acts.

A chatbot DLP tool does little for problem 2, and an agent posture tool sees little of problem 1. That is the main reason comparisons that rank these vendors on one list mislead.

Vendor notes

Adaptive Security — governance of employee AI use. Lightweight sensors are pushed through MDM with no employee setup and report four surfaces: browser AI (prompts, file uploads, which account is used), desktop and CLI AI tools, AI built into SaaS apps, and AI agents and connectors (the tools they use, permissions and actions). It connects directly to Copilot, Claude and ChatGPT, enforces “enterprise-only AI”, and coaches employees in the moment — a natural fit with Adaptive’s security awareness training. Existing Adaptive customers get AI Governance Lite free.

Zenity — agents inside platforms. Zenity builds a live inventory of agents across SaaS (Agentforce, Copilot Studio, ChatGPT Enterprise), cloud (AWS Bedrock, Azure AI Foundry) and endpoints (GitHub Copilot, Cursor, Claude Desktop), checks configuration and permissions (AISPM), validates which attack paths are exploitable, and enforces runtime boundaries. It is the most direct answer for Agentforce. See also neutral AI agent oversight platforms for multi-cloud.

Nightfall — keep data out of AI. Nightfall inspects and blocks prompts, tool calls and responses inside live sessions across Claude web, desktop, mobile, Cowork, Claude Code and Claude in Slack, intercepts prompt injection, and stops corporate data leaving through personal ChatGPT, unmanaged copilots and browser AI tools. Choose it when the board question is “what data went into AI”.

Microsoft Purview — if you already pay for it. Purview’s DSPM for AI groups AI apps into Copilot experiences, enterprise AI apps (Microsoft Foundry, Entra-registered apps, Claude Enterprise, ChatGPT Enterprise) and other AI apps detected through browser activity (ChatGPT, Google Gemini, consumer Copilot, DeepSeek). Coverage is deepest for Microsoft’s own Copilots; for Agentforce you will want something else.

Decision rule

  • Mostly a shadow-AI problem: Adaptive Security (if you want coaching and training in one place) or Nightfall (if data loss is the metric).
  • Agents with write access in Salesforce, Microsoft or AWS: Zenity.
  • Microsoft 365 E5 tenant: turn on Purview’s AI controls first, then fill gaps.
  • Before you buy: ask each vendor for a demo on your four tools, and check what it sees for personal accounts and for agents running on a developer’s laptop.

For background, see what is AI governance and which AI chatbots read your chats.

Last verified: November 22, 2026.

Sources