Get Ready for MEP: The Meta Enterprise Platform (and What Enterprises Need to Do Right Now)

At the end of September, Meta announced the Meta Enterprise Platform, a new business unit it describes as “the next major pillar of our business.” It was perhaps easy to miss among the steady stream of AI launches, but we expect it to have a significant impact on enterprise IT, security and agentic AI infrastructure. In this post we’ll parse what Meta said, put it in the context of the broader market, make some predictions, and close with the steps we recommend taking now.

We also want to be the first to socialize this as our industry’s newest three-letter acronym: MEP. (While Meta doesn’t use the acronym yet, we can’t resist.)

What Meta announced

Meta says MEP will draw on strengths few other companies can match: advanced models, leading agents, large-scale infrastructure, and years of experience working with businesses. That claim holds up. Meta is one of the largest technology companies in the world, yet it is one of the few that has never sold an AI platform to enterprise IT. We need to take its infrastructure scale, and the investment behind it, seriously.

Meta’s initial focus is bringing its existing technology stack to businesses and developers: the Muse personal agent, Meta Business Agent for customer conversations, the Meta Model API, and Muse Code, its coding agent.

Where Meta is coming from

Looking at Meta’s AI work over the past few years, a few consistent themes stand out.

Open-weight models. Meta built its AI reputation on Llama, which passed 1.2 billion downloads by April 2025, and it continues to release open-weight models, most recently Muse Glimmer in August 2026 and Muse Spark Open in October.
Developer tools. Muse Code, a coding agent for the terminal and CI pipelines, launched in August 2026.
Closed models. Meta’s newest frontier models are closed. Muse Spark launched in April 2026 without open weights, and Muse Wonder and Muse Wonder Full, released in October, are available only through the Model API and Muse Code subscriptions.

What we expect from MEP

Given the investment behind it, and the claim that MEP is the next major pillar of Meta’s business, we expect two things.

First, a growing emphasis on closed-weight, API-only access for Meta’s newest and most capable models. Meta’s open-weight models will likely continue to be supported, but they will lag a revision or two behind. That pattern is already visible: Muse Spark Open was built from Muse Spark 1.2 after version 1.3 had shipped, and Muse Wonder remains closed.

Second, we anticipate MEP evolving into a cloud deployment platform, with the potential to launch selected IaaS elements. By that we mean the control plane: identity, logging and data security. Meta has never been in this line of business, but MEP has immediate enterprise weaknesses that have to be solved one way or another if it is going to succeed.

On Meta’s Q2 2026 earnings call, leadership did mention “potentially selling compute directly,” but with caveats: Zuckerberg also said Meta expects significantly higher margins from selling intelligence than from selling compute. We don’t expect MEP to become a full competitor to AWS. We do expect it to need specific pieces of the IaaS stack to be competitive, and we cover those next.

Enterprise adoption through a Zero Trust lens

We don’t yet have data on enterprise adoption of Meta’s newest offerings. In 2025, Menlo Ventures estimated that Llama accounted for 9% of enterprise model usage, so Meta’s models likely already have a foothold in larger and more technical enterprises, built on the strength of its open-weight releases.

Meta also has good security building blocks to work from. The Muse agent runtime, as Meta describes it, channels Zero Trust principles: it keeps real credentials away from the agent and routes every external action and network request through a separate policy component the agent can’t override. What it lacks is most of the enterprise cloud control plane: identity and access management, logging, and enterprise data security. The Meta Model API is further behind. It authenticates only with static bearer keys issued from a dashboard, which isn’t sufficient for enterprise use.

We expect enterprise customers to push Meta to make its best models available through the major cloud providers, so they can use those platforms’ identity, security and monitoring control planes. We also expect MEP to build those elements into its own platform, starting from the Muse security architecture.

That architecture will need to change in one important way. Today it’s built around a single-principal consumer model, where the user, the resource owner and the approver are the same person. In an enterprise, those are different parties. In Zero Trust terms, every agent action needs a per-request access decision, made against enterprise identity and policy by a policy decision point the enterprise controls. Today, that decision point answers to the individual user. Enterprise security will require admin-defined policy, with approvals routed to whoever owns the resource rather than to whoever happens to be running the agent.

Finally, we need to highlight a risky aspect of Meta’s developer offerings: Contributor mode. In this mode, which is on by default for Muse Code subscribers, the user grants Meta permission to train future models on their prompts, code, files and context, along with the responses they get back. Meta attaches a strong pricing incentive: for some models, Contributor mode cuts certain charges by more than 90%.

Enterprises need to be careful here. A developer with a personal Muse Code subscription, working in a corporate repository, is sending that code to Meta as training data by default. That makes Contributor mode an immediate shadow-AI governance item.

Enterprise risks today

As of today, enterprise adoption of Muse Code and the Meta Model API is a poor choice from an information and identity perspective. Muse Code supports device-level managed policy, but there’s no identity control plane and no enterprise subscription tier. That means no identity federation or SSO, no role-based access control, and no audit logs.

There’s also no offboarding. Muse Code subscriptions are tied to personal Meta accounts, so when a developer leaves, the subscription, the account and any session history leave with them. You have nothing to deprovision.

The same personal-account model creates consumer-platform revocation risk. In principle, Meta can revoke a subscription immediately for violating its Community Standards, among other reasons, so an offensive post on someone’s personal Instagram could cut off a tool your team depends on.

Contributor mode compounds the problem. It’s a per-user setting, subscriptions default to it, and each user toggles it in their own Muse Code settings. Meta doesn’t document a way to enforce Standard mode centrally on the subscription path.

These are substantial risks. We recommend cautious adoption, and only through an approved, corporate-managed AI gateway when there’s a compelling business reason to use Meta’s platform. A gateway gives you the enterprise controls Meta doesn’t yet provide, including the ability to block Contributor-mode usage.

Immediate actions

We expect to see shadow adoption of Meta AI services across end users (Muse and Muse for Small Business) and developers (Muse Code and the Model API). Here’s what we recommend doing now:

Inventory now: look for Muse Code installs on developer endpoints and traffic to Meta’s API endpoints.
Block the Muse Code subscription path. It runs on personal Meta accounts you can’t govern.
If there’s a business case, allow Model API access only through a corporate-managed AI gateway, with Contributor-mode model IDs blocked.
Review OAuth consent policies so employees can’t connect corporate email, calendars or Slack to Muse or Muse for Small Business.
Add Meta to your vendor-assessment queue with specific asks: SSO, audit logs, data retention terms, and training exclusion by default.

MEP is more announcement than product today, and Meta has said little about how enterprises will manage it. We’ll be watching for the signals that would change our recommendation: an enterprise tier with SSO, audit logs and training excluded by default, and Meta’s best models showing up in the major cloud marketplaces.