Arcjet vs Jamf AI Governance

Arcjet vs Jamf AI Governance

Jamf AI Governance is part of Jamf for Mac. It inventories the AI apps, CLI agents, and MCP servers running on managed Macs, and it deploys each AI tool's own enterprise settings as Jamf blueprints that users can't edit. Arcjet records each connection that a macOS or Windows device makes to an AI service, attributes it to a person, and blocks the providers you don't approve. Arcjet also decides each tool call, prompt, or model switch in the hooks of Claude Code, GitHub Copilot, Cursor, OpenAI Codex, and Muse Code.

The two products answer different questions, and they work together. The Arcjet endpoint agent deploys through Jamf, and Arcjet can map local Mac accounts to Jamf users. If you already run Jamf for Mac on your fleet, turning on AI Governance needs no extra license and gives you vendor-correct settings for Claude Code and Codex, so start there. Add Arcjet when you need Windows devices, coding agents that Jamf doesn't configure yet, a policy decision on each agent action, or a record of which person reached which AI provider.

Arcjet publishes this comparison. Competitor details come from Jamf's public pages, press releases, and documentation, reviewed on November 3, 2026. Several Jamf capabilities described here are announced with general availability targeted for Q4 2026, and products change, so check the linked sources before you decide.

Arcjet vs Jamf AI Governance at a glance

Jamf AI Governance configures AI tools through the Mac management plane, and Arcjet observes AI connections and decides agent actions. The following table compares the two products across 12 areas.

What is Jamf AI Governance?

Jamf AI Governance became generally available on June 30, 2026 as a capability of Jamf for Mac. Jamf's pricing page says it comes built into Jamf for Mac and Jamf for Mac – Higher Education, which list at $12.50 per Mac per month, billed annually, with a 25-device minimum. Jamf lists Jamf Pro, Jamf Connect, and Jamf Protect as the products behind Jamf for Mac. The AI Governance page says it isn't available to K-12 institutions.

Visibility comes from Jamf Protect. Jamf's launch post says that without the Jamf Protect agent deployed across your fleet, "AI Governance has no signal to work with." You can deploy Jamf Protect only to feed AI Governance, and the press release says no other agent is required.

Jamf builds two inventories from that telemetry. The AI Application Inventory lists AI apps, CLI harnesses, and tool invocations across the fleet, and it flags higher-risk activity such as SSH, osascript, and credential access. The MCP Server Inventory shows which MCP servers are running, what they expose, and which AI clients connect to them.

The policy builder exposes the enterprise settings that each AI vendor ships: model selection, file system access, MCP server connections, and tool usage, and the press release adds tenancy and network permissions. Jamf's brochure describes three postures, Permissive, Balanced, and Strict, validated by Anthropic. You publish a policy as a Jamf blueprint, which uses Apple's Declarative Device Management, and it lands on the Mac as a managed settings file that the user and local processes can't edit, remove, or replace. Jamf's Terraform provider guide says the settings are "the tool vendor's own configuration format, not Jamf's," validated against a schema for each tool. The press release says policies take effect offline and before a user's first login to an AI agent.

The launch post names Claude Code and Claude Cowork, alongside OpenAI Codex on AWS Bedrock, and the press release adds Claude Desktop. The post says Cursor and GitHub Copilot will follow as their enterprise controls mature. The launch post lists three prerequisites: Jamf Pro connected to Jamf Account through OIDC single sign-on, blueprint privileges (Create, Read, Update, and Delete), and the Jamf Protect agent.

AI Governance produces an Executive AI Posture Report as a PDF with an EU AI Act crosswalk, and the press release describes SIEM integration and registering discovered agents with Okta for AI Agents.

At JNUC on September 23, 2026, Jamf announced investigation of flagged AI activity down to the device, user, and command line, estimated AI usage and spend by user without capturing prompts, personal-account visibility and policy in Safari, Chrome, Edge, and Firefox, and AI governance for mobile devices. Each has general availability targeted for Q4 2026.

What is Arcjet?

Arcjet is an AI agent runtime security platform with an endpoint agent for shadow AI on macOS and Windows, and policy enforcement in the hooks that coding agents already fire.

The endpoint agent observes each connection from a managed device to an AI service, identifies the signed app that made it, and attributes it to the person behind the local account at that moment. On macOS, it's a Network Extension content filter, with a root launchd service that owns reporting and credentials, so closing the menu bar app or logging out doesn't stop managed reporting. On Windows, it's a service that runs under LocalSystem from boot. It reads the destination server name from the TLS ClientHello through Windows' built-in Packet Monitor API and correlates kernel network and process ETW events to identify the process and user, with no kernel driver. Neither version decrypts TLS, installs a root certificate, or reads prompts, so pinned apps and CLI agents keep working. You deploy the agent through your MDM, including Jamf, and devices enroll with short-lived tokens or device credentials.

In the Arcjet Console, Discovery goes from People to Devices to AI usage, and a Providers view shows which people, devices, and apps reach each service. Counts are connections, not prompts or tokens. Endpoint provider policies set Allow or Block per provider for a group of devices. The agent blocks the connection on the device, and each decision lands in Arcjet Activity with the person, device, app, and provider. Detection rules ship as a reviewed manifest that updates without a new agent release.

Arcjet coding agent hooks enforce one policy across Claude Code, GitHub Copilot, Cursor, OpenAI Codex, and Muse Code. You install the hooks through managed settings, MDM, or an organization lock. Each tool call, prompt, or model switch is decided before it runs, and Arcjet records the session. Starter policies include MCP allowlists, egress allowlists, model allowlists, credential access, and sensitive information, and evaluation runs at the edge.

Arcjet also records activity without enforcing it, from the Claude Compliance API and from OpenTelemetry over HTTP at https://decide.arcjet.com. For custom agents, JavaScript, TypeScript, Python, and Go SDKs provide guard() and protect().

For endpoint agent pricing and availability, book a demo. Platform plans are Individual at $25 per month, Startup at $299 per month, and custom Enterprise pricing.

How do Jamf AI Governance and Arcjet differ?

Jamf AI Governance and Arcjet differ in what each observes, how each ties activity to a person, what each controls, how each blocks a service, and which agents and operating systems each covers. The following sections cover each difference, then the areas where Jamf goes further.

Process inventory compared with connection records

Jamf starts from processes. Jamf Protect sees which AI tools run on a Mac, what launched them, and which tool invocations carry risk, such as an agent that runs osascript or reads a credential store, and Jamf's brochure says the inventory also shows what each tool connects to. A process-level signal can also see work that never reaches the network, such as a model running in Ollama or a stdio MCP server that talks to its agent through a pipe. Arcjet's endpoint agent can't, because those make no connection.

Arcjet starts from connections. The endpoint agent records that claude, signed by Anthropic, reached api.anthropic.com from a specific Mac, for a specific person, at a specific time, and that a browser on the same Mac reached chatgpt.com. Because the record starts from the provider, it includes AI services reached from a browser or from an app that isn't on any list of AI tools, and the same record works on Windows. A connection from an AI app isn't proof of an AI request, because it can be an update check, and Arcjet counts connections, not prompts or tokens.

Neither signal is complete on its own. For coding agents, Arcjet treats the endpoint connection, the hook event, and the Claude Compliance API as three sources that each prove something different. A hook event near a connection isn't proof that both belong to the same request. For more information, see shadow AI in coding agents. If you want an inventory before you buy anything, a Jamf Pro advanced computer search with the Application Title criterion finds installed AI apps, as how to detect shadow AI describes.

Attribution to a person

Jamf ties AI Governance to people through single sign-on for administrators and through blueprint scoping, so a policy applies to the Macs of a group. Jamf has announced investigation down to the user and command line, and usage by user, both targeted for Q4 2026.

Arcjet attributes each connection to a person. An administrator opens a device in Discovery and maps each local account to a person: an Arcjet team member, a person already in Discovery, or a synced Jamf user. One person can have accounts on several devices, and an assignment takes effect from a date the administrator picks. Shared, system, and unidentified accounts show as Unknown user and don't inherit the device owner. On a build Mac that three engineers sign in to, Arcjet doesn't attribute their activity to whoever Jamf lists as the assigned user.

Vendor settings compared with a decision on each action

Jamf deploys the settings that each AI vendor already supports, and the AI tool enforces them. For Claude Code, those are keys such as availableModels, permission rules, and MCP restrictions, delivered as managed settings that outrank every user and project setting. Because the tool applies the settings itself, they work offline and before the first login. The cost is that each vendor's settings cover different things. Jamf Protect reports the tool invocations it observes, but the allow or deny comes from the tool's own settings, and Jamf's documentation doesn't describe a record that ties each invocation to the setting that allowed or refused it.

Arcjet decides each action in the agent's hook. When Claude Code is about to run cat ~/.aws/credentials, call an MCP server that isn't on your allowlist, or send a request to a host outside your egress allowlist, the hook asks Arcjet first, and the policy allows or denies the call before it runs. The same policy applies in Cursor, GitHub Copilot, Codex, and Muse Code, and each decision lands in Arcjet with the session, the rule, and the arguments. Each decision is a request to Arcjet at the edge, while managed settings apply on the device with no network call.

Blocking an AI service

Jamf's launch post says AI Governance "handles the part that comes after yes" and that if you want to block an unsanctioned tool, "Jamf already does that elsewhere." In Jamf Pro, restricted software matches an exact process name and can kill the process when it starts.

Arcjet blocks at the provider. An endpoint provider policy sets Block for a provider on a group of devices, the agent refuses the connection on the device, and the attempt appears in Arcjet Activity with the person, device, app, and provider. Arcjet doesn't need to read HTTPS content to do this.

A hostname can't tell a personal account from a company account on the same service. A developer on a staff MacBook who runs claude with a personal Max plan reaches the same api.anthropic.com as a colleague in your Claude Enterprise organization. A vendor tenancy setting is the right control for that case, and Jamf lists tenancy among the settings it deploys. For Claude Code, the managed settings forceLoginMethod and forceLoginOrgUUID restrict login to your organization. For the controls on other vendors, see blocking personal AI accounts on work laptops.

Agents and operating systems

Jamf AI Governance runs on Macs and configures Claude Code, Claude Desktop, Claude Cowork, and OpenAI Codex, with Cursor and GitHub Copilot planned. An engineering team that standardizes on Cursor, or a Windows group that uses GitHub Copilot, sits outside its coverage until Jamf adds those tools.

Arcjet's endpoint agent runs on macOS and Windows, and its hooks cover Claude Code, GitHub Copilot, Cursor, OpenAI Codex, and Muse Code on macOS, Linux, and Windows. Neither product sees a personal phone or a device without its agent. Cloud agent sessions, such as Codex cloud tasks or the GitHub Copilot cloud agent, run on the vendor's infrastructure, so an endpoint agent sees none of their traffic. Arcjet hooks reach some of those sessions where the vendor loads repository or organization configuration.

Where Jamf goes further

Jamf AI Governance does several things that Arcjet doesn't. It costs nothing extra on Jamf for Mac and runs on the agent and console that your Mac team already operates. Its MCP Server Inventory shows what each running server exposes, and its process telemetry flags invocations such as osascript whichever agent made them. It writes each vendor's settings from presets validated by Anthropic and tracks vendor releases for changed controls, and its tenancy settings separate personal from company accounts, which a connection record can't. The posture report, its EU AI Act crosswalk, and the Okta for AI Agents integration have no Arcjet equivalent.

When to choose Jamf AI Governance

Choose Jamf AI Governance when your fleet is Macs managed with Jamf Pro and Jamf Protect, and your AI tools are the ones it configures. Jamf AI Governance fits when any of the following apply:

• You already license Jamf for Mac, so AI Governance needs no new contract.
• Your developers use Claude Code, Claude Desktop, Claude Cowork, or OpenAI Codex.
• You want each vendor's enterprise settings, such as model limits and login tenancy, deployed as tamper-proof managed settings.
• You want an inventory of running MCP servers and what they expose.
• You need a posture report for auditors or the board.
• Your organization doesn't run Windows devices for the people who use AI tools.

When to choose Arcjet

Choose Arcjet when you need to know which person reached which AI provider, on macOS and Windows, or you need a policy decision on each coding-agent action. Arcjet fits when any of the following apply:

• You manage Windows devices as well as Macs, or your Macs aren't on Jamf Pro and Jamf Protect.
• Your developers use Cursor, GitHub Copilot, or Muse Code alongside Claude Code and Codex.
• You want one policy that decides each tool call, prompt, or model switch across five coding agents, with a record of each decision.
• You want each AI connection attributed to a named person, with shared accounts kept separate.
• You want to block unapproved AI providers on the device without decrypting TLS.
• You want Claude Compliance API and OpenTelemetry records in the same console as endpoint and hook evidence.

Can you use Jamf AI Governance and Arcjet together?

Yes. Jamf configures the AI tools on your Macs, and Arcjet records which person reached which provider and decides each coding-agent action. Arcjet already works with Jamf in two places.

First, you deploy the signed Arcjet endpoint agent package to your Macs through Jamf, the same way you deploy other software. Second, you can connect a Jamf Cloud tenant to Arcjet in your organization's Discovery settings. You create a read-only Jamf Pro API role and client with the privileges for computer inventory and users, and Arcjet syncs them on a schedule. The Jamf credentials never reach the Mac. When you assign people to Mac accounts, the Person menu lists Arcjet team members, Discovery people, and synced Jamf users in one list, and you can attach the Jamf computer record as device evidence. The administrator still confirms who uses each account. Arcjet connects one Jamf tenant per organization, and the connection supports Jamf Cloud tenants such as https://example.jamfcloud.com, not self-hosted Jamf Pro.

If Jamf AI Governance and Arcjet both deliver Claude Code managed settings, check which one Claude Code applies. By default, Claude Code uses only the highest-ranked managed source that sets a policy key, ranks an MDM profile above a managed-settings.json file, and skips the rest without a warning. Run /status in Claude Code on a test Mac: it names the managed source in use and the sources it skipped. To apply both, put the Arcjet hook configuration in the same source as your Jamf policy, or set managedSourcesBehavior to "merge" in the highest-ranked source, which requires Claude Code v2.1.242 or later.

What are the alternatives to Jamf AI Governance?

The alternatives to Jamf AI Governance, or products to run beside it, differ in where they look and which operating systems they cover. The following products cover overlapping ground:

• Arcjet records AI connections on macOS and Windows, attributes them to people, blocks unapproved providers, and decides coding-agent actions through hooks in Claude Code, GitHub Copilot, Cursor, OpenAI Codex, and Muse Code.
• Microsoft Defender for Endpoint lists local AI agents and their configured MCP servers on Windows and macOS devices onboarded to Defender. For more information, see Arcjet vs Microsoft Agent 365.
• Runlayer's AI Watch reads configuration files and processes to inventory AI clients, MCP servers, skills, and plugins, and installs hooks. For more information, see Arcjet vs Runlayer.
• Prompt Security, part of SentinelOne, inspects employee AI use in the browser and redacts data. For more information, see Arcjet vs Prompt Security.

For a wider comparison of shadow AI tools, including SSE, browser, and endpoint products, see shadow AI detection tools.