Privacy check · filed 2026-10-03 · last checked 2026-10-02

Six AI girlfriend apps promise end-to-end encrypted chats. Their own privacy policies don't.

No AI companion app we track encrypts chats end to end, and none can: the AI that writes the replies runs on the company's servers, so the company has to be able to read what you send. Six apps say otherwise on their own sites as of October 2, 2026: OurDream, Secrets, GoLove, Dondi, Xotic and CrushOn. Not one of their privacy policies repeats the claim. The strongest says chats are encrypted in transit and at rest, and four of the six let your chats be used to train AI models. Two more apps, Candy and Lovescape, made the same promise and took it down in August and September.

We earn affiliate commission from five of these six apps (OurDream, Secrets, GoLove, Dondi and Xotic, through CrakRevenue) and from Candy. It did not change what we report. How we work.

Why a chatbot can't be end-to-end encrypted

End-to-end encryption means only the two ends of a conversation hold the keys. The company in the middle relays your words but cannot read them; that is how Signal and WhatsApp work. In an AI companion app the other end of the conversation is the company itself: a language model on its servers, or a provider it pays, has to read your message in plain text to write the reply.

Nomi, which does not make the claim, says so on its own wiki: 'End to end encryption is not possible with an AI chatbot as we would need to have the keys to decrypt the message'. What these apps can honestly offer is encryption in transit (HTTPS between you and them) and at rest (on their disks). Both are worth having. Neither stops the company, its moderators or its AI providers from reading a chat.

The six claims, against each company’s own privacy policy

App	What its site says	What its privacy policy says	Claim on site
OurDream	'Your chats are safe with end-to-end encryption.' 'No one can read them, not even us.' (homepage FAQ)	Encrypted 'both at rest and in transit using industry-standard AES encryption and TLS 1.3 protocols'. Its own privacy summary says AI providers read chats: 'The message you send is processed by the model provider to generate the reply'.	Every snapshot since July 6
Secrets	'Judgement-free conversations protected with end-to-end encryption' (homepage)	'Encryption at rest: AES-256 encryption applied to all stored personal data and content', plus TLS in transit. Data is used to 'train and improve our AI models'.	Every snapshot since July 7
GoLove AI	'Every chat uses end-to-end encryption.' 'Private conversations are not stored on internal servers' (homepage)	No encryption claim at all, only 'appropriate security measures'. Stored Content includes 'the discussions with the AI Companions', kept 'for a period of 6 (six) years after you cease being a User'.	Every snapshot since August 3, our first
Dondi AI	'All conversations are end-to-end encrypted, your data is never sold or shared' (homepage FAQ)	Lists 'encryption' among generic security measures. 'Your Content and interaction data may be processed using AI systems and related technologies operated by us or our technology partners', including for 'AI model training and improvement'.	Every snapshot since August 3, our first
Xotic AI	'We encrypt everything end-to-end, so your discussion never goes anywhere it shouldn't.' (homepage)	Never mentions encryption. Collects 'messages and chat history' and shares information with 'third-party service providers who perform services on our behalf'.	Every snapshot since August 3, our first
CrushOn	'Your conversations and fantasies stay completely private with end-to-end encryption.' (its /ai-sex-chat landing page)	Never mentions encryption, and says it 'cannot guarantee that any information, during transmission or while stored on our systems, will be absolutely safe from intrusion by others'. 'We may use User Content from character chats to train AI models.'	First captured September 27; still up October 2

Details worth knowing

OurDream contradicts itself most directly. Its homepage says no one can read your chats, 'not even us'. Its own Trust & Safety privacy summary has a table of who sees private chats, and two rows besides you say yes or sometimes: staff 'Only with your explicit support consent, or when responding to a lawful legal request', and third-party AI providers, 'Yes'.

GoLove says private conversations 'are not stored on internal servers'. Its privacy policy, from 404 Intelligence LTD, counts your chats as stored Content and keeps personal information for six years after you close the account.

Dondi's privacy answer reads like a copy of the one Candy deleted. Candy's homepage used to say it 'uses end-to-end encryption to keep all AI GF conversations safe and confidential. You have full control over your chat history and can enable two-factor authentication for extra security.' Dondi's says 'you have complete control over your chat history at all times. Two-factor authentication is also available for additional account security.' Candy took its version down; Dondi's is still up.

CrushOn puts the promise on a marketing page, not its homepage, and its community guidelines say people read chats: 'We use a combination of human moderators and automated systems to monitor and review Crushon accounts, characters, and interactions'.

Two apps took the claim down

Candy made the claim in every snapshot we took from July 6 to September 11. Between September 16 and 22 the answer was rewritten to 'Candy AI uses industry-standard security measures, including encrypting data in transit using TLS 1.2+ protocols.' The full before-and-after is in our changelog record.

Lovescape's homepage had a section headed 'End-to-End Encryption Everywhere' with the line 'Fully encrypted, with no sharing or logging', from our first snapshot on July 6 to August 21. Even then its own fine print described ordinary encryption: chats 'encrypted in transit using industry-standard TLS 1.3+ and at rest with AES-256 encryption'. By September 11 the section was gone, and the page now advises: 'Avoid entering unnecessary personal details in prompts or chat.'

What the others say

Nomi: no end-to-end encryption, and it explains why (quoted above).
Kindroid: chats encrypted in transit and at rest, and 'We will only decrypt content for human review or disclosure where you ask us to (for example, in a moderation appeal)'. The closest anyone comes to the promise, and still not end-to-end.
Janitor AI: 'We use HTTPS to encrypt information in transit', and nothing more is claimed.
CamSoda AI: 'Conversations are encrypted in transit and at rest.'
FantasyGF: 'end-to-end privacy in all forms of communication'. It sounds like encryption and promises nothing anyone can check.

What to assume

Treat a companion chat like an email to the company. Its staff can read it under the company's own rules, so can the AI provider that writes the replies, and it is kept for as long as the privacy policy allows. Leave out real names, addresses, workplaces and photos of real people.

How we checked

We read each app's homepage and privacy policy from a US address on October 2, 2026, and compared them with our tracker's dated snapshots of the same pages, which go back to July 6 for most apps (August 3 for GoLove, Dondi and Xotic, when we started tracking them). Every quoted phrase on this page is checked against the saved page text before it is published. We list an app as making the claim only when its own site says chats or conversations are end-to-end encrypted.

Sources

To cite this page: The Eliza Report, "Six AI girlfriend apps promise end-to-end encrypted chats. Their own privacy policies don't.", filed 2026-10-03, last checked 2026-10-02, https://elizareport.com/privacy/end-to-end-encryption/