Really? Are we really going to trust him again?

Are we going to allow Meta CEO Mark Zuckerberg, whose company just agreed to pay a settlement worth up to $17.1 billion over 10 years for its addictive design and harm done to kids’ mental health, to sucker us into handing over more control and more online information in the form of signing up for Meta’s new AI agent, Muse?

Yes, I think we might.

Meta has done an amazing job of making a powerful agent running on generative AI feel cute, fun, and easy to use. Muse works so well, people are likely not going to pay attention to the warning signs.

That’s the problem.

The danger is that Muse is so slick that it’s easy to let a powerful AI — one that will make mistakes because all generative AI makes mistakes — act on your behalf and stick you with the consequences.

At a time when we need to be more aware of what our tools do for us, when more powerful AI models are being released every week, and in the absence of any sign of regulatory oversight or guardrails, ceding control is the most dangerous thing we can do.

And although Meta makes it very easy to let AI take the wheel, there’s a way to flip the script.

Ride Down the Slippery Slope

After having my blood boil watching Zuckerberg shill Muse at a launch event, I resolved to use it myself. I needed to find out first hand if my anger was misplaced or, as I assumed, there was cause for concern.

My concern was justified.

Starting at https://muse.ai/, I was presented with a login screen which included my Facebook and Instagram accounts. Setup required I step through a couple screens before being presented with a question: what would I like to call it? It suggested some cute names. Then the avatar sat there, silently, blinking, waiting for me to proceed.

The inclination of most people will be to pick a name for Muse. Something fun. And that’s the wrong decision.

Muse is a thing. It is an agentic system designed to follow its instructions, to act on your behalf, and to, ultimately, make Meta more money. It is a cute, fuzzy face slapped across a multinational mega-corporation.

💡 First tip: Don’t give it a name. Just as most farmers don’t name animals they’re going to eat, you need to always remember you’re in control. You make the decisions because you’re the one responsible for them.

Why am I giving you tips instead of telling you to never use Muse?

Because there are legitimate use cases for this thing which I’ll describe.

So I want to help you protect yourself and anyone you know who uses Muse, while arming you with the rationale for the recommendations.

Peeking Behind The Curtain

How Muse works is based on a set of files you can access by clicking “Library” in the sidebar then selecting “System Files” at the bottom of the screen. The one you need to edit first is named SOUL.md.

Yes. Soul.

Explanatory text describes what is contained there:

About this file. This is Muse’s persona. It sets out the values and habits Muse tries to hold to in every conversation. It starts from a template that ships with Muse. Muse may refine it over time and tells you when it does. You can edit it at any time. This note is not part of the file.

And the default first line you will see in the file is, “You’re not a chatbot. You’re becoming someone.”

Creepy. As. Hell. My skin is crawling as I type this.

💡 Second tip: change the first line to something like, “You’re a capable tool working for [your_name]. Be direct, skip the personality, and flag risks plainly.”

You need to be the one in control here. Change its core construct and establish it will work on your behalf.

There is one more setting you have to change before going any further. It lives in the Settings menu under “Data Controls.”

Labeled “Help improve our AI models” and toggled on by default, this allows Meta to use everything you do with Muse to train its AI. From the very beginning, you hand over everything you do in Muse to Meta which enables it to continue harming kids and monetizing divisiveness.

No thank you.

💡 Third tip: turn off the model training toggle. You will have to confirm your choice — which neither OpenAI nor Anthropic require you to do in their platforms — but the desired state is only two clicks away.

What Is It Good For?

I and other people I talk with struggle to use Meta’s platforms. Their settings and permissions are arcane, twisting labyrinths, so Muse can be an effective tool for navigating them.

Facebook Marketplace postings leave me cold. I am stymied by the process of describing the thing I want to sell and setting a price for it.

In my test, Muse made the process very simple. I gave it three images, described an issue with the item, and it spun up a Marketplace listing with recommended pricing which I could review and modify with it.

And when I tested posting permissions, they worked as described in the terms of service. Muse made me confirm the listing before it would post it.

That’s good, safety-first design and genuinely helpful.

Other uses could include managing an organization Facebook Page and other things which Meta seems to make harder than they should be. For those uses, Muse is worthwhile.

And they’re the only things I will use Muse to do. I do not trust Meta or Zuckerberg.

You should also be wary, because Muse does come with costs and risks you need to know about.

Read The Fine Print

You are responsible for everything Muse does on your behalf. Every post. Every action. Whether it asks you or its settings allow it to act for you. Regardless, once it acts, you own the outcome.

You are on the hook for almost everything it does for you. Meta’s liability is capped at $250, and arbitration applies unless you opt out by physical mail.

From the moment you create an account, a heartbeat — yeah, that’s its term sitting alongside “Soul” — runs every 30 minutes. On a free account you’re not paying for it, but somebody sure is paying for the electricity and compute for that machine activity.

So, every single Muse account, unless you turn off the heartbeat, adds to the generative AI compute burden driving the planning and construction of data centers.

💡 Fourth tip: tell it to turn off its heartbeat. Honest. Just tell it, in a prompt.

Once it tells you it’s done, confirm it by clicking on the avatar, then selecting “Upcoming” so you can avoid inadvertently adding more AI activity. It’s the smart call for you and the planet.

Learn The Lessons

Those four tips were selected purposefully.

The recommendations are straightforward ways you can take back control from Muse and Meta. In doing them, you intentionally re-wire how Muse works.

People are going to use Muse. Your kids or partner may already be using Muse right now.

As AI tools get more capable and more cuddly — and that damn avatar, which you can’t turn off on desktop, is cute — the real danger is in using them ignorantly.

Don’t blindly accept the tool you’re handed. Flip the script. Tell it how you want it to work for you.

Because when you do that, you have control and not the guy smirking on stage.