The Permissions Problem

Why Generic Access Controls Are a Liability in the Age of AI

The music industry runs on information asymmetry. A label executive knows things the agent doesn't. The agent knows things the promoter doesn't. A manager knows things that no one but the artist should know. This asymmetry isn't dysfunction — it's architecture. It's how deals get made, how leverage is maintained, how trust is earned and protected over years of working relationships. When we are talking about business & team operations, the daily work of getting things done inside the business to the benefit of our artists, permissions are a critical point because collaboration is the name of the game.

Now, introduce an AI layer. Specifically, introduce an MCP — a Model Context Protocol server that gives a language model live access to your operational data via your software’s APIs. Suddenly, that carefully maintained asymmetry is only as strong as your permission logic. And if your permission logic is "admin, user, guest," you don't have architecture. You have a liability.

Generic permissions were built for a different threat model.

Admin, user, guest — these tiers were designed for human operators navigating a UI. A human user clicks around, gets redirected when they hit a wall, maybe sees a "403 Forbidden" and moves on. The surface area of what they can access is bounded by what they think to look for, how much time they have, and the interface's own guardrails.

An LLM doesn't work that way. When you give a language model access to a platform via an MCP, you're not giving it a UI — you're giving it an API surface. It doesn't browse. It queries. And it queries with a thoroughness no human user would ever apply in practice. Ask it to pull a tour summary, and depending on how your permissions are structured, it may also surface the guaranteed minimums, production costs, advance balances, and the settlement details — not because you asked, and not because the model is malicious, but because all of that data lives in the same permission tier and the model is doing its job.

Generic access controls create a flat data surface.

Flat data surfaces leak.

The cracks are structural, not incidental.

This isn't a bug in any particular tool. It's an architectural mismatch between how generic SaaS permissions were designed and how AI agents actually interact with data. The assumption baked into "user-level access" is that a user will only touch what they need because they're a person with a specific task and limited time. That assumption breaks entirely when the "user" is an LLM.

Sensitive artist data requires customizable access permissions

The standard for AI-era permission architecture isn't generic role-based access control with three tiers. It's granular, configurable, context-aware permissioning — where access can be scoped not just to a role but to a specific data type, artist, relationship, or workflow context.

Take Airtable, for instance: you can grant or deny access to a specific base, but you cannot grant or deny access to specific pieces of information within that base, and then scope that further to only specific artist information in that base. Most SaaS companies skimp on this in favor of broader feature sets, but AI breaks those models. Everyone who builds software knows how difficult it is to rewire a system’s entire permissions architecture when it’s not built in from the start, so they don’t do it.

This level of control isn't a luxury feature. In an MCP-connected environment, it's the baseline for responsible deployment.

If you're running basic permissions, you are high risk.

The tools that haven't been built for this problem will tell you their existing permission model is sufficient. It isn't — not when the access layer is an AI agent. The gap between "sufficient for a human navigating a UI" and "sufficient for an LLM traversing an API" is exactly where sensitive data leaks through.

Artist data is not generic business data. It carries legal weight, contractual sensitivity, artist location and financial information, and personal trust that took years to build. The infrastructure protecting it needs to match the stakes.

Generic permissions were fine for a different era. That era is over.

Wishing you an inspiring weekend!

Matt Urmy

CEO, Artist Growth