Is Muah AI safe?

I do not recommend Muah AI. In September 2024 the site was breached, and about 1.9 million email addresses were exposed together with the prompts people had written, according to Have I Been Pwned. Its published policies do not describe what changed afterwards, so this page explains what was reported, what it means for you, and where to go instead.

This is a cautionary page, not a review. I did not sign up for Muah AI, and I am not linking to it.

The short answer

No, not on the evidence available. A breach can happen to any company. What makes this one different is the kind of data involved, the way it was tied to real identities, and the lack of any published incident report or policy update I could find afterwards.

If you used Muah AI before late 2024, assume your email address and prompts may be in that data. The steps further down tell you what to do about it.

What happened in 2024

Here is the breach as described by the sources that reported it. I have stuck to what they said.

Have I Been Pwned

Have I Been Pwned (HIBP), the breach-notification service run by security researcher Troy Hunt, lists the Muah.AI breach as follows:

• Breach date: September 2024
• Added to HIBP: Oct 8, 2024
• Accounts: 1.9 million email addresses
• Data exposed: AI prompts, email addresses and sexual fetishes

HIBP says many prompts were sexually explicit and some described child exploitation. Because of that, it classed the breach as sensitive. Sensitive breaches are not publicly searchable: you can only check by signing in to HIBP with the email address itself, or as a domain owner.

404 Media

The breach was first reported by 404 Media in an article by Joseph Cox published on Oct 8, 2024. According to that report:

• The hacked database contained users’ interactions with their chatbots.
• Prompts were linked to email addresses, many of which appeared to be personal accounts with real names.
• Many prompts described child sexual abuse scenarios.
• The hacker said they found vulnerabilities relatively quickly and described the site as a handful of open-source projects put together loosely.

Other coverage

Other outlets added detail:

• CyberInsider (Oct 9, 2024) gave the breach date as Sep 17, 2024. It reported Troy Hunt’s comments that most of the addresses were not disposable and could be tied to real identities, and that because the site verified emails, the owners of those addresses were the ones using the accounts. The site’s administrator blamed a targeted attack by competitors.
• Malwarebytes (Oct 9, 2024, updated Oct 11) said Muah.AI had claimed its communications were encrypted and that it did not sell user data, and that the leaked data contradicted those claims. Its update noted reports that the data was being used in extortion attempts, without confirming the details.
• Linklaters, the law firm, wrote on Oct 14, 2024 that reports described threat actors contacting IT employees found in the data to seek access to their employers’ systems.

The extortion reports are secondhand. I have not seen them independently confirmed, and you should read them as “reported”, not “proven”.

Why this breach is worse than most

Most breaches leak an email and a password hash. That is bad, and you change your password. This one leaked something you cannot change: what you wrote.

Three things combine here:

1. Explicit content. The exposed data was prompts on an adult platform, in many cases sexual.
2. Real identities. Per Troy Hunt, most addresses were not throwaway accounts and were verified by their owners.
3. No undo. Once a database like this is copied, it does not get un-copied.

That is the recipe for blackmail, which is exactly what the later reports were about. It is also why I keep saying the same thing across this site: use a separate email for adult services, and never write anything that identifies you.

What Muah AI’s policies say

Here is what its official pages said when read on Oct 25, 2026. A second check the same day was blocked, so I cannot rule out a newer version.

Company

The terms name only “MUAH.AI” with no further legal entity. They are governed by US law, and the site footer says “California Based”. There is no company name, address or registration number that I could find. If you needed to complain to a regulator, you would not know where to start.

Privacy policy

The privacy policy I read is marked effective 5-17-2022, which is before the breach. It covers name, email, phone, usage data and cookies, used for providing the service, personalisation, marketing emails and analytics. It does not specifically address chat content or whether it is used to train AI.

I found no official incident report, breach notice or policy update describing what Muah AI changed after 2024.

Age checks

The terms say you must be 18 or older, and the privacy policy says accounts proven to be under 18 are suspended. No age-verification method is described.

Real people

I found no rule in the terms on real people or deepfakes. The user guide only says to use adult fictional characters and images you have permission to use. Meanwhile, its character creator works from an uploaded image. Under this site’s rules, a service without a written ban on real-person content does not get recommended, breach or not.

Payments and refunds

Payment methods and the processor are not stated on the public pages. The terms say subscriptions are not refundable, though support may grant a one-time refund in good faith. Plans run up to $999.00 a year for the top tier, as shown in the upgrade screen on Oct 25, 2026.

Apps

I found no current App Store or Google Play listing; the store links checked returned errors. Android is offered only as a direct APK download from the site. Installing an APK from a company that has already been breached is not a step I would take.

Were you affected?

If you ever signed up, do this:

1. Check HIBP. Because this breach is flagged sensitive, you need to verify the email address with HIBP before it shows you the result.
2. Delete what you can. Muah AI’s guide describes a “Wipe Cloud Data” option and a “Delete Account” flow, and the privacy policy says deletion requests go to its support email. This removes what is on its servers now. It cannot touch data already leaked.
3. Change reused passwords. If the password you used there is used anywhere else, change it there.
4. Expect phishing. Treat any message that mentions Muah AI, your chats or “your fantasies” as a scam until proven otherwise.
5. Do not pay anyone. If someone threatens to expose you, do not pay and do not reply. Keep the messages and report them to the police. Paying does not guarantee it ends.
6. Tell IT if a work address is involved. The Linklaters report was specifically about attackers targeting employees. Your employer would rather hear it from you.

Safer options

None of these is “breach-proof”, and I have not paid for any of them yet. But each one names a company, publishes its rules, and bans content that depicts real people.

Secrets AI

Run by Secret Labs Inc., which its terms describe as registered in both Canada and the United States, with a Delaware address. Its prohibited content policy bans sexual content depicting identifiable real people without verified consent, including deepfakes, and it does not support image uploads at all. Card statements show “S LABS INC”, per its wiki. The free tier is 20 messages a day; Premium is $19.99 a month. See the Secrets AI test drive.

Candy.ai

Run by EverAI Limited in Malta, which means EU privacy law applies. Its community guidelines ban images that resemble real persons and non-consensual intimate imagery. Charges show as “EverAI” on your statement. Prices are listed in EUR and shown as discounts, from EUR 3.99 a month billed annually. See the Candy.ai test drive.

OurDream AI

Run by Dream Studio USA, Inc. in Delaware. It bans content that resembles real individuals, including celebrities, and does not allow reference photo uploads because of deepfake risk. Where the law requires, a third-party provider checks age and the company receives only the outcome. See the OurDream AI test drive.

GirlfriendGPT

Run by NextDay AI Incorporated in Montreal, the same company behind SpicyChat. Its guidelines ban impersonation and uploading images of others without consent. Payments go through CCBill or TrustPay, per its terms. See the GirlfriendGPT test drive.

For more, the best AI girlfriend apps page lists everything I cover, and how I test explains the difference between a test drive and a crash test.

How to stay safer on any AI companion site

• Use a separate email that does not contain your name and is not linked to work or social accounts.
• Do not sign in with Google, Apple or Facebook if you want the account kept apart from your identity.
• Never put your real name, workplace, city or photos into chats or characters.
• Use a unique password and turn on two-factor sign-in where offered.
• Prefer services that publish a company name, a privacy policy with a recent date and a rule against real-person content.
• Avoid installing APKs from sites you do not already trust.