Is Meta Muse HIPAA Compliant? (What Doctors Need to Know)

Short answer: no. Meta’s Muse is a consumer personal agent launched September 8, 2026. Meta does not offer a Business Associate Agreement for it, model training on your conversations is on by default, and Meta itself says the stronger “Confidential VM” isolation is still coming. A medical practice cannot route protected health information (PHI) through Muse today.

What Muse is

Muse is Meta’s personal AI agent from Meta Superintelligence Labs, launched September 8, 2026 on iOS, Android, and the web, with WhatsApp as a messaging channel. It runs in a “Secure VM” with its own browser, a shell, persistent files, sub-agents, reusable skills, and cron scheduling. It connects to Gmail, Google and Outlook services, Calendly, Instagram, Facebook, and more, and it can shop, book travel, fill in forms, and build documents. Setup takes about a minute.

Muse is free with a weekly token limit, with Power and Maximum tiers at roughly $20 and $100 per month on iOS (less on the web). It runs only on Meta’s own model, Muse Spark. Model training on your conversations is turned on by default and can be switched off in settings. At launch it was available in the United States and Canada only.

What HIPAA actually requires

Under HIPAA, a medical practice (a “covered entity”) can only hand protected health information (PHI) to a vendor that has signed a Business Associate Agreement (BAA). A BAA is the contract where the vendor promises to safeguard PHI, report breaches, and accept HIPAA obligations. No BAA means no PHI, full stop, no matter how good the product is.

A consumer privacy promise is not a BAA. Even a strong one. The BAA is what makes the vendor legally accountable to your practice, and it has to cover the whole chain: the app, the model, and the servers.

Five reasons Muse fails the test today

• No BAA. Meta has published no HIPAA statement or BAA for Muse.
• Training on by default. Your conversations improve Meta’s models unless you opt out, which is incompatible with PHI.
• Meta can access the current VM. Meta clarified that the “Meta cannot see inside” claim applies to a future Confidential VM, not the Secure VM Muse runs on now.
• One personal identity. Muse is tied to a Meta account and consumer connectors like Instagram and Facebook. A practice needs staff roles and an audit trail.
• Early over-sharing reports. Launch-week press carried reports of Muse sharing personal details it should not have, which Meta said it was looking into. Agent mistakes with PHI are reportable breaches.

The HealthEx connector that lets a person link their own medical records to Muse does not change any of this. A patient choosing to share their own data is allowed under HIPAA; a provider sharing patient data with Meta is not, without a BAA.

The HIPAA alternative that works the same way

What makes Muse attractive is the feel of it: an agent that remembers you, talks to you on WhatsApp, runs jobs in the background, and needs no setup. PhiClaw is that feel built for medical practices.

Here is the part most buyers miss: PhiClaw is the same kind of product. It is the HIPAA build of OpenClaw, the open-source agent that Grok Bot and Muse are so often compared to. So you get the same shape of experience:

• Always on. PhiClaw runs in the cloud 24/7, answers patients at 2 a.m., and keeps working when the office is closed.
• Message it like a coworker. Talk to it on WhatsApp, iMessage, Slack, Telegram, the web, or the iOS and Android apps.
• Memory and routines. It remembers your practice, your preferences, and your patients’ context, and it runs scheduled jobs such as no-show follow-ups, recall reminders, and weekly reports.
• Does multi-step work. Intake, scheduling, insurance questions, SOAP-note scribing, e-prescribing, billing, fax, website updates, and social posts, end to end.
• Sub-agents. A front-desk agent, a back-office agent, and a marketing agent coordinate under one roof.

What changes is everything around the agent. PhiClaw signs a BAA with every practice on every plan. It runs on HIPAA-eligible AWS (including Amazon Bedrock for the model) and Convex, both under BAA, with calls through Retell AI, also under BAA. Your data is never used to train a model. PHI stays inside the HIPAA-compliant PhiClaw app, encrypted and audit-logged, while consumer chat apps are used only for non-PHI convenience. And it ships with its own HIPAA EHR and CRM, plus connections to 30+ major EHRs and 300+ integrations through Keragon.

Checklist before using any AI agent with patients

• Get a BAA signed for the exact product.
• Confirm every subprocessor (cloud host, model provider, telephony) is also under BAA.
• Confirm your data is never used for model training.
• Make sure PHI is encrypted, audit-logged, and tied to staff roles, not one personal login.
• Keep a licensed clinician as the decision-maker on anything clinical.

Frequently asked questions

Does Meta sign a BAA for Muse?

No. As of October 2026 Meta has not published a HIPAA statement or a Business Associate Agreement for Muse.

If I turn off training, is Muse HIPAA compliant?

No. Turning off training removes one problem, but HIPAA compliance requires a signed BAA, BAA-covered infrastructure, access controls, and audit logging. Muse offers none of those for a practice.

Can my patients use Muse to manage their own health?

Patients can share their own records with whatever tool they like; HIPAA governs providers and their vendors, not individuals. The restriction is on your practice sending PHI to Meta.

What does PhiClaw do differently?

PhiClaw signs a BAA with every practice on every plan, runs on HIPAA-eligible AWS and Convex under BAA, never trains on your data, keeps PHI inside its encrypted and audit-logged app, and ships with an EHR, CRM, intake, scribing, e-prescribing, billing, fax, and patient messaging.

How much does PhiClaw cost?

PhiClaw is flat and predictable: Starter at $300/mo for a single clinic, or Enterprise at 30% of what we save you, with a BAA on every plan.