Is Candy AI Safe? Is It Legit? Privacy Check

Candy AI is a real business, not a scam. It is operated by EverAI Limited, a company registered in Malta, and it delivers what it sells. It is not a private service, though: no independent privacy researcher has ever audited it, its age check is a checkbox, and its own affiliate program placed ads on a deepfake porn site in 2025.

That is the short answer, and the two halves of it usually get argued as if they were the same question. "Is Candy AI legit" and "is Candy AI safe" are different things. One is about whether you get charged for nothing. The other is about what happens to several months of unusually personal conversation sitting on someone else's server. Below is what has actually been documented, with dates and sources, and what you can do about the parts that are in your control.

Is Candy AI legit, or is it a scam?

It is legitimate. Candy AI launched in 2023, has operated continuously since, and provides the chat, image generation, voice and video features it advertises. Independent reviews through 2025 and 2026 consistently reach the same conclusion: whatever its faults, taking money and disappearing is not one of them.

The complaints that do repeat are billing complaints, and they are worth understanding because they get mistaken for fraud. Public reviews cluster around three things: confusion over what a token actually costs, subscriptions that proved harder to cancel than expected, and refund requests that went nowhere. None of that is theft. It is the predictable friction of a product that sells you a subscription and then meters the features you actually came for, which is a business model designed to make your bill unpredictable rather than a scheme designed to defraud you. If you want the full arithmetic on that, we broke down what Candy AI really costs once tokens are included separately.

Who owns Candy AI?

EverAI Limited, a company registered in Malta. This is stated in Candy AI's own terms of service and was independently confirmed by the investigative outlet Bellingcat in January 2025. The platform is headquartered in Santa Venera, Malta.

Knowing the operator matters more in this category than in most. A named, registered European entity is meaningfully better than the anonymous operators behind a lot of AI companion sites, because it means there is somebody to write to, somebody a regulator can act against, and a jurisdiction whose data protection law applies. Malta is in the EU, so GDPR covers the company's processing. That does not make the service private. It makes the service accountable, which is a lower bar and still a real one.

What did Bellingcat find about Candy AI?

On 28 January 2025, Bellingcat published an investigation into the commercial relationships around MrDeepFakes, a site that distributed non consensual deepfake pornography and had accumulated close to two billion content views. Candy.ai advertising was among what the investigation found running there, placed through Candy AI's affiliate program, with pop up ad copy that was crude enough that Bellingcat quoted it directly.

EverAI's response was that it does not condone or promote the creation of deepfakes, and that the placements had been made by affiliate webmasters running hundreds of sites the company had not individually vetted. The links to Candy.ai were removed from MrDeepFakes after Bellingcat asked about them.

Read that for exactly what it is. It is not evidence that Candy AI makes deepfakes, and it would be dishonest to present it that way. It is evidence that the company runs an aggressive affiliate program, pays commission on recurring revenue, and was not checking where its brand ended up. Affiliate networks are genuinely hard to police, which is why it is now fairly routine for brands to see where their ads are actually running rather than trust a partner's word for it. Candy AI evidently was not doing that in 2025. Whether that changes your decision is a judgment call, but it belongs in the decision.

Has Candy AI been independently audited for privacy?

No, and that absence is the most useful thing to know here. The best known independent assessment of this category is the Mozilla Foundation's February 2024 Privacy Not Included review, which examined 11 romantic AI chatbots: Eva.AI, Romantic AI, iGirl, Anima, Genesia, Chai, Talkie Soulful AI, CrushOn.AI, Mimico and Replika. Candy AI was not among them.

You will find pages claiming Mozilla flagged Candy AI. They are wrong, and we had that error in our own notes before checking. What Mozilla actually found is still relevant as context, because the findings were about the category rather than any one app. All 11 received the warning label, which Mozilla called one of the worst product categories it had ever reviewed. Ten of the 11 failed its minimum security standards. More than half would not let users delete their data. Seventy three percent published nothing about how they handle security vulnerabilities.

So the honest position on Candy AI is that nobody credible has looked. In a category where independent review found near universal failure, an unreviewed app is not a safe app. It is an unknown one, and you should price it accordingly.

What does Candy AI do with your chats?

Your conversations are stored on EverAI's servers, and authorized staff can access them for content moderation, abuse prevention and debugging. That is not unusual and it is not hidden. Every serious app in this category does some version of it, because a platform generating explicit content has a legal obligation to police what happens on it.

The practical consequence is the same one that applies to Replika, to us, and to everything else here: treat anything you type as something that exists permanently, on a server you do not control, tied to your account and your payment method. That is a genuinely different exposure from a text message to a friend, and the intimacy of the format makes it easy to forget. People tell these apps things they have told nobody else. The comparison with Replika is instructive, because Replika did get audited and the findings were bad; we covered what the privacy research actually found about Replika in detail.

Two things are worth doing regardless of which app you choose. Use an email address that is not your main one, and never reuse a password. If a companion app is ever breached, the damage is not the password. It is the association between your real identity and the contents of the chat log.

Is Candy AI safe for minors?

No, and the safeguards are thin. Access to adult content is gated by a self declaration that you are over 18, with no identity check, no document upload and no biometric step. A thirteen year old clicks the same box an adult clicks.

The regulatory environment has moved on this. New York's AI companion law took effect on 5 November 2025 and requires AI disclosure at the start of a session and at least every three hours during one. California's SB 243 took effect on 1 January 2026 and adds disclosure notices, crisis protocols and break reminders for minors. These rules apply to companion services generally, and enforcement across offshore operators is still an open question. If you are a parent, the checkbox is the whole gate, and device level controls are more reliable than anything the platform is doing.

Is Candy AI trustworthy enough to pay for?

For most adults, yes, with two specific cautions. The company is real, the product works, and the failure mode people actually experience is a bill larger than they expected rather than anything sinister. If you want the full picture on the product itself, our honest Candy AI review goes through what it does well and where it falls down.

The first caution is financial. Understand the token meter before you subscribe, because the advertised price is a floor. The second is informational. There is no independent verification of how this company handles your data, and given what independent review found everywhere else it looked in this category, absence of evidence is not comfort. Where rivals do publish their terms, they differ a great deal on who ends up owning what you write, which we compared across the best AI companion apps for privacy.

If either caution matters enough to change your mind, the alternative is not to go without. It is to pick something with a flat price and fewer unknowns. We built a Candy AI alternative with no tokens and one flat price for exactly that reason: $9 a month billed annually, nothing metered, so a heavy month and a light month cost the same and there is no meter running while you talk.

How to use an AI companion app more safely

None of this requires giving up the product. It requires treating it like what it is, which is a service you are renting from a company, rather than a private space.

Use a dedicated email address so a breach cannot be linked back to your main identity. Set a unique password, since weak passwords were one of the specific failures Mozilla found across this category. Pay through a method you can cancel independently, such as an app store subscription, rather than handing a card directly to a site with a documented pattern of refund complaints. Assume every message could be read by a person, and calibrate what you share to that assumption rather than to how private the conversation feels. Check the cancellation path before you subscribe, not after you want out. And if you are going to buy annual, buy it after a month of use, not on day one, because the thing most people discover in week three is not a privacy problem. It is that the product was not what they thought they were buying.