Is Candy AI safe?

Candy AI is an adults-only companion service that, according to its terms, is run by EverAI Limited of Malta, and whether it is safe for you depends mostly on what you are willing to put into its chats. Its privacy notice, revised July 30, 2026, says outside AI providers may receive the content of your messages and that de-identified interactions can be used to train its models. Its product page says your messages are not used for training. Its terms, revised August 17, 2026, license your input to Candy for training its AI models, so the three documents give different answers to the question most people ask first. Candy’s help center says there is no native iPhone app, and the App Store listing called “Virtual Chat with Candy AI” has a seller we could not connect to Candy.

We could not open candy.ai from our own tools on October 16, 2026, so we read cached copies of its terms, privacy notice, underage policy, community guidelines, blocked content policy and product page through a content service. Each copy carries the revision date we give, and a newer version may exist. We read Candy’s help center in full through its public API, checked Apple’s store data for the listing, and used two outside sources, a data-practice grade and a 2025 investigation, each labelled as secondary where it appears. We did not use Candy AI ourselves. Riho publishes this page and makes a competing app, so weigh it that way. In the sources we checked we found no regulator action, fine or lawsuit involving Candy.

Candy AI safety at a glance

Is Candy AI legit?

According to its terms, Candy AI is owned and operated by EverAI Limited.

Candy’s terms and its privacy notice both name EverAI Limited as the company behind it, say it is incorporated in Malta, and give a street address in Santa Venera and the registry number C107181. We did not check that number against the Maltese registry, so it is Candy’s own statement. The privacy notice also says Candy has appointed a Data Protection Officer and a UK representative and gives a privacy contact, and the product page promises 24/7 support through its help center. Card statements show the charge as “EverAI”, according to the help center, which is worth knowing so you recognize it later.

A registered company is a starting point and says little about safety. Whether it is safe to use depends on what the company’s own documents say about your data, your age, your money and the content, and the sections below go through each.

Candy AI privacy policy: what it collects and shares

The privacy notice, revised July 30, 2026, is organized as a table of purposes, the data used for each, and the legal basis. For an account it lists your email address (a mandatory field), an encrypted password, a nickname, your gender, and your name if you sign in through a service such as Google. It also lists the date and provider of your signup and the results of any third-party age estimation or identity check. For running the service it lists your currency and the country and city it detects from your IP address, your token balance, sign-in dates and IP addresses, and your preferences for the companion you chat with.

Your messages are covered too. The notice lists the prompts you enter and the output generated in response, including calls, “to the extent these contain Personal Data”. It adds that your gender, combined with your preferences about companions, may count as sensitive personal data under data protection law, and that Candy will ask for consent where that applies. Device and browser details, support messages and cookies are on the list, and so are payment records, which are handled by outside processors and include the card brand and last four digits, bank details or a crypto wallet address depending on how you pay.

Who receives your chats

Section 5 of the notice says Candy relies on outside service providers for payments, hosting, email marketing, affiliate tools, and moderation and AI tools, “including at our discretion third-party LLM providers and/or hosters”. It then says plainly that these third parties may receive the content of your messages. It does not name the providers. It also says sharing may involve transfers outside the European Union, including to the United States, and the notice offers a copy of the safeguards for those transfers on request.

On sale of data, the notice says Candy will not disclose, sell, trade or otherwise transfer your personal data to third parties without your consent where required, or unless the notice says otherwise. One of the things it says otherwise is that in a restructuring, sale or change of control your data may go to a successor or buyer, and that it stays subject to the notice, as updated. Security gets one sentence: “appropriate security measures”. The product page says conversations are encrypted, and the notice names no specific control, so the encryption claim rests on the marketing page alone.

Does Candy AI use your chats for training?

Three documents answer this question differently.

The product page, in the cached copy we read, says your conversations are “encrypted, never shared” and that Candy doesn’t use your messages for training. Its FAQ is narrower: chats are “not used to train public models or shared with third parties”. Those two sentences can be read as different statements, since a private model trained on your chats is not a public one.

The privacy notice lists, as a purpose, research to “train and develop our AI models and moderation technologies” and to “prepare datasets for further training”. The data for that purpose is described as information associated with you, including exchanges with companions, that may be aggregated, anonymized or de-identified, and the notice says dataset preparation may include human review of de-identified interactions. A separate quality-assurance purpose says content data is randomly queried to check that generation tools work.

The terms, in section 4, are the broadest. You keep ownership of your input, and you grant Candy a perpetual, worldwide, sub-licensable license to use it for purposes that include “training, developing, enhancing, evolving and improving” its AI models and moderation tools, and for marketing and promotion, in a manner consistent with the privacy notice. We found no opt-out from training in any of the documents, and none that treats free and paid users differently. VerifyWise, an outside index that scores the disclosure in public documents, flags both this training clause and the perpetual license as dealbreakers.

If you read only the product page you would conclude your chats are never used. If you read the notice and the terms you would conclude de-identified versions of them may be. The terms describe themselves as legally binding and the notice as the statement of how data is processed, so weigh them above the product page, and keep anything you would not want in a training set out of the chat.

Candy AI data retention and deletion

Section 6 of the privacy notice sets the periods. The table gives them in one place.

The terms add that Candy may remove account content after defined periods of inactivity, without a refund or restored credits. The terms therefore allow account content to disappear sooner than the notice’s three years.

How to delete your account, and what a reset does

Candy’s help center gives the path: My Profile, Settings, the Danger Zone, then Delete Account. It says this makes your preferences, usage history and personal information permanently unavailable, apart from data Candy must keep for legally required periods, and that deleting before a renewal date also cancels the subscription. Cancelling alone does not delete anything, according to the help center. Only deleting the account removes your content.

Chats work differently from accounts. The help center says Candy has no option to clear a chat’s history while fully keeping the AI’s memory, that Reset Chat starts the conversation fresh and Delete Chat removes the thread for good, and that the reset and delete options clear the conversation and its memory together. A deleted conversation cannot be recovered, and relationship progress and streaks usually cannot be restored. If a conversation matters to you, copy what you want before you reset it.

The privacy notice lists rights to access your data, correct it, restrict or object to its use, receive it in a structured machine-readable format, and ask for erasure “in limited circumstances”. Requests go to the privacy contact in the notice, and Candy says it tries to answer within one month. We found no self-service export button in the help center. The documents do not say whether a deletion request reaches the outside providers that received your messages.

Candy AI age rating and age rules

The notice says Candy provides no services to anyone under 18, or under the age of majority where the user lives, and that it will delete information and block a user it learns is underage. The community guidelines repeat the 18 requirement. The product is built for adults: Candy’s help center lists 18+ images among its paid features and describes Live Action as a Premium-only erotic video feature on select characters.

How the age is checked is the weaker part of the picture. The Underage Policy, revised October 10, 2025, describes an age gate on entering the website, where users must affirm they are over 18, and a disclaimer that giving a false age breaks the terms. It says Candy “may also implement further measures” to verify adults “in accordance with local requirements”, and it says the company can’t be held responsible for inaccurate age statements. The privacy notice supports the idea of stronger checks in some places, since it lists the results of third-party age estimation or identity verification as data collected “as applicable based on your location”, and the terms say Candy may ask you to verify or re-verify that you are an adult. None of the documents says which countries get which check. We did not test the gate.

Is Candy AI safe for kids? No. It is an adult product by its own description, and the checks we could read rely mostly on what a user declares. Parents who find it on a child’s device can use device-level controls and a direct conversation with the young person.

Correction, October 16, 2026: an earlier version of this page cited an App Store age rating of 4+ as Candy’s own. That label belongs to a listing whose ownership we cannot confirm, which the next section covers, so it no longer appears as a Candy age signal here.

Candy AI on the App Store: is the listing official?

Candy’s help center answers the iPhone question directly. The article, last updated June 19, 2026, says Candy.ai has no native app for iOS or Android. You use it in a mobile browser, or add the site to your home screen from Safari as a web app. Our Candy AI review goes through that route.

An App Store listing called “Virtual Chat with Candy AI” exists in the US store, and several things about it differ from what Candy’s own pages say about its product. The seller is V-LOST O.E., and Apple’s data shows it was released on May 27, 2026 and updated to version 1.0.1 on July 23, 2026, with four ratings averaging four stars. Its privacy policy and terms of use are linked as Google Docs. Its description offers free access to three messages a day with one character, a different free allowance from the five messages in total that Candy’s help center and terms describe (Candy’s product page also advertises a seven-day trial). The description names “Candy AI” but never mentions the candy.ai site or EverAI, and the listing is rated 4+, English only. A search of the US store for “candy ai” also returns other apps with similar names from other sellers, and none of the results lists EverAI Limited as the seller.

The listing may still be Candy’s. What we can say is that nothing we found ties it to Candy, and Candy’s help center does not mention it. The terms say Candy disclaims liability for costs arising from use of “a third-party distribution platform that is not indicated or recognized by us”. That clause is general and does not name this listing. Our position matches the position in our Candy AI review, where the listing’s ownership is also treated as unconfirmed. The practical steps are short. Before you sign in or pay in an app with Candy’s name, check who the seller is, check that Candy’s help center or site points to it, and if in doubt ask Candy’s support at the address in its terms.

Payments, refunds and cancellation

We describe payments in general terms here and leave current prices to Candy’s checkout page. The terms say paid plans are monthly, quarterly or annual, renew automatically at the price shown at checkout, and that price rises are communicated before a renewal. Paid plans include unlimited messaging and a monthly token amount shown at checkout, generally 100 according to the terms, and tokens pay for extras such as images and voice. Payment methods vary by location. The terms mention credit cards and pay-by-bank, and the help center adds crypto, Apple Pay, Google Pay and bank transfer. Our review lists the secondary price figures we found.

Refunds are narrow. The terms and the help center agree that you must ask within 24 hours of payment, that a request fails if you have used more than 20 tokens on a subscription, and that only card purchases qualify. A cancellation is not a refund request. Users in the EU and UK have a separate 14-day withdrawal right on the initial purchase, and it does not apply to renewals. A chargeback cancels the purchase at once and revokes the tokens, and moderation actions and removed characters are not refunded.

One conflict is worth knowing about. The help center says cancelling your subscription does not delete your tokens and that they stay on your account, hidden while you are on the free service. The terms, revised August 17, 2026, say remaining tokens expire and become unusable when paid access ends, while the help center article, updated August 25, 2026, says they stay. The two conflict, so use your tokens before the paid period ends.

Content rules, moderation and human review

The Community Guidelines, revised March 6, 2026, and the Blocked Content Policy, revised July 30, 2026, list what is prohibited. The lists cover content involving or resembling minors, non-consensual intimate imagery and deepfakes, images that resemble real people, impersonation of celebrities, hate and discrimination, illegal content, instructions for violence, drug material, jailbreaking the moderation controls, and material related to self-harm. The Blocked Content Policy describes the service as private chats and generated images visible to the individual user, and says users can’t upload or share content on it. Adult images are allowed within the guidelines, according to the help center.

Moderation is automated first. The Underage Policy says Candy’s controls run on its own LLM technology, and the terms say they may change a request before an output is produced if the request seeks prohibited content. When the controls flag something, the terms say Candy “or authorized partners may access and manually review the flagged content and/or other content associated with the account”. That can end in removed content, a terminated account or a report to the authorities, and Candy reports known child sexual abuse material to the National Center for Missing and Exploited Children. Chats are “generally private”, and the guidelines add that Candy may monitor communications for quality assurance. The help center says a message flagged by mistake can be reported with the in-product report option and that its Moderation Team reviews reports. It does not publish response times or how many reports lead to action.

The terms also set expectations about what the product is. They say it is for entertainment only and not emotional support, that exchanges with characters are fictional, that characters can’t keep real-world promises and that offers to meet should not be taken seriously, and that users should not disclose financial details, addresses, contact information or passwords in chats. We found no self-harm protocol or crisis resource in the documents we read, only the guideline that prohibits self-harm material and the terms’ advice to contact a qualified professional if you are distressed. If you are struggling, you can call or text 988, the US Suicide and Crisis Lifeline, or contact your local helpline elsewhere.

Press and outside reports

The Bellingcat piece is about an advertising relationship, and Candy’s response was that the placement ran through its affiliate programme, which offers partners commission on subscriptions and token purchases. Candy’s current guidelines and blocked content policy prohibit deepfakes and images that resemble real people, which is the policy position. We have no test of how well it is enforced. Our search on October 16, 2026 found no fine, lawsuit or regulator action against Candy or EverAI. A single search can miss things, so check the regulator and court records for your own country if the question matters to you.

What we could not confirm

We read cached copies, not live pages, so a revision after those dates would not appear here. We could not read Candy’s subscriptions page and give no current prices. We did not use Candy, so we did not test the age gate, moderation, deletion or its memory. The documents don’t name the outside AI providers that receive chats, say whether deletion reaches them, say which countries get an age check beyond the declaration, or publish moderation statistics. The product page’s claim of encryption has no matching detail in the notice. The help center and the terms disagree about tokens after cancellation. VerifyWise lists the policy date as August 17, 2026, which matches the terms and not the notice, so we treat the notice’s July 30 date as the one we read.

A checklist before you use Candy AI

How Riho compares

Candy AI offers many characters, with generated images, video and a token balance. Riho is one girlfriend, made for adults and on iPhone. In the app you can read what she remembers about you and read her diary, and the memory page explains how.

Her privacy page, revised October 16, 2026, says outside AI providers receive only the context a feature needs and that requests for access, a copy, correction or deletion go by email to privacy@rihoapp.com. The questions in the checklist above apply to Riho as much as to Candy, so read that page against the same ones about training, retention and deletion before you rely on any app, ours included. The safety checklist for AI girlfriends walks through them in order. For the products side by side, see Riho vs Candy AI, or Candy AI alternatives if you are leaving. The best AI girlfriend apps list gives sourced facts for each app it covers. Early access to Riho opens from the button below.

Frequently asked questions

Is Candy AI safe to use?

It is an adults-only service run by a company that says it is registered in Malta, with published terms, a privacy notice revised July 30, 2026 and a help center. It can pass your message content to outside AI providers, can use de-identified interactions to train its models, generally keeps account data for three years after your last activity, and has no opt-out from training that we could find. Use it as an adult, keep identifying details out of chats, and read the notice before you pay.

Is Candy AI legit?

Candy AI is operated by EverAI Limited, which its own documents say is registered in Malta under number C107181, and it has a help center, published policies and a data protection officer contact. Those facts show it publishes the documents and contacts a company would publish. Whether it is safe to share intimate details with is a separate question, answered by what the privacy notice says it does with your chats.

Is Candy AI 18+, and is it safe for kids?

Candy’s notice says the service is for adults, meaning 18 or the age of majority where you live, and its paid features include adult images and erotic video. Its Underage Policy describes an age gate where users affirm they are over 18, with further checks where local rules require them. It is not a service for children, and the checks we could read rely mostly on what the user declares.

Does Candy AI use my chats for training, and how long does it keep them?

The product page says messages are not used for training, while the privacy notice says aggregated, anonymized or de-identified interactions may be used to train and develop models, and the Terms license your input for training. Account data is generally kept for three years after your last activity unless you ask for deletion, financial records for ten years, and interaction logs are deleted after 30 days.

Is the Candy AI app on the App Store official?

We could not confirm it. Candy’s help center says Candy.ai has no native iOS or Android app. The US listing called Virtual Chat with Candy AI names V-LOST O.E. as the seller, links Google Docs for its policies and does not mention Candy or EverAI, so treat its 4+ rating as that listing’s own and check with Candy before you sign in or pay there.