首頁 / ShortFling 部落格 / Safety
本文目前僅提供英文版。閱讀英文版 →
SafetyAI Companion Apps and Privacy: What Leaks Teach Us
Why this keeps coming up
People tell AI companions things they would not tell a friend. That is part of the appeal, and it is also the problem: a companion app is a database of confessions, flirtation and sometimes photos, run by a company you probably know nothing about. Over the past year, security researchers have shown several times how easily that database can end up open to anyone.
This post walks through the main incidents reported between October 2025 and March 2026, what they have in common, and a checklist you can use on any companion or adult platform, including ours. It is not a security audit of any specific app.
Three incidents worth knowing
Chattee Chat and GiMe Chat, October 2025. According to Biometric Update’s summary of the research, these two AI companion apps exposed 43 million messages and 600,000 photos from more than 400,000 users through an unprotected server.
Chat & Ask AI, February 2026. Malwarebytes reported on February 9, 2026 that a researcher found about 300 million messages tied to more than 25 million users of this app, developed by Codeway, readable through a misconfigured Firebase database, a well-documented type of error that leaves backend data accessible without authentication. Full chat histories, model choices and settings were exposed, along with data from other Codeway apps. Codeway fixed the problem across its apps within hours of being told, according to the report. The same researcher reportedly found the same type of flaw in 103 of 200 iOS apps scanned.
The Oversecured audit, March 2026. Biometric Update reported on March 19, 2026 that the mobile security firm Oversecured examined 17 popular Android AI companion apps, including AI girlfriend and boyfriend apps, with more than 150 million combined installs. It reported 14 critical and 311 high-severity flaws. Ten of the 17 had flaws that could reach users’ conversation histories. The problems included cloud credentials hardcoded into an app and a flaw in another that could expose chat databases, cached photos and voice messages.
What these leaks have in common
None of them involved a sophisticated attack. They came from basic configuration mistakes: an unprotected server, a misconfigured database, secret keys shipped inside the app. That matters for how you judge risk. A big brand name is no guarantee, but a small app that grew fast and ships a new character every week is exactly the kind of product where basics get skipped.
Regulators have mostly looked elsewhere. Italy’s data protection authority fined Replika’s developer, Luka Inc., 5 million euros in a decision reached on April 10, 2025, citing a missing legal basis for processing, an inadequate privacy notice and weak age verification. In September 2025, the US Federal Trade Commission opened a study of seven companies offering companion-style chatbots, including Character Technologies, Meta, OpenAI and Snap, asking among other things how they use or share personal information from users’ conversations. Oversecured argues that no regulator has yet acted against a companion app specifically over app-level security flaws. Treat that as a reason to protect yourself rather than wait.
A privacy checklist for any companion or NSFW app
Assume anything you type could one day be read by a stranger. That is not paranoia after the incidents above; it is a planning assumption. Write as if a leak would be embarrassing but not damaging.
Keep your real identity out of it. Malwarebytes’ advice after the Chat & Ask AI leak was to avoid using your real identity when discussing sensitive subjects and to avoid uploading personal documents. Do not give a chatbot your full name, workplace, address or the names of people in your life.
Think twice about photos. Images are the hardest thing to take back, and two of the incidents above involved cached or stored photos. Never upload anyone else’s picture; besides the privacy risk, sharing intimate images of another person without consent is a crime in many countries.
Read the training and sharing settings. Many AI products let you opt out of having your content used to improve models. Look for that setting when you sign up, and check again after major updates, because defaults can change.
Use a separate email and a unique password, and sign in with an account you can afford to lose. If an app offers a way to delete your history or your account, find out where it is before you need it.
Prefer apps with a real privacy policy. You want a named company, a stated jurisdiction, a list of the kinds of providers that get your data, and a way to request a copy or deletion of it. If you want your conversations as a keepsake, see how to back them up rather than trusting the app to keep them safe.
What this means for readers and writers
The risk in a companion app comes from what you put into it. The more of your life a product asks for, through confessions, memories or selfies, the more there is to lose. Reading a story asks for much less, because the story is written for everyone and does not need anything about you. Writing your own fiction sits in between: a draft about invented adult characters is far less sensitive than a diary-like chat, as long as you keep real people out of it. If you want the companion experience in a different form, our guide to AI boyfriend apps versus story romance compares the two.
What is true about ShortFling
ShortFling is not an AI chat or companion app, so there is no stream of personal conversations stored with us. You can browse the fiction library and read free chapters without creating an account. Entering the 18+ NSFW area, including the curated adult short stories, means confirming you are an adult, and that confirmation is stored in a cookie. Depending on your region, stricter checks may apply.
If you sign up, for example to use the writing tool, our privacy policy lists what we collect, such as your email, purchase records and reading activity, says we do not sell personal data, and names the kinds of service providers that process it, including the AI text providers behind the editor. Signed-in users can request account deletion from their account page. We are not going to claim more than that: no app should promise it can never be breached, and you should apply the checklist above to us too.
This is a general overview, not legal or security advice.
FAQ
Have AI companion chats really leaked?
Yes. Reported cases include Chattee Chat and GiMe Chat in October 2025 (43 million messages and 600,000 photos) and Chat & Ask AI in February 2026 (about 300 million messages from 25 million users), both caused by unprotected databases.
How can I use an AI companion more safely?
Keep your real name and identifying details out of chats, avoid uploading photos, use a separate email and unique password, opt out of model training where possible, and check how to delete your data before you need to.
Does reading on ShortFling require sharing personal chats?
No. ShortFling is a fiction platform, not a companion app. You can read free chapters without an account; signing up is needed for features like writing, and our privacy policy describes what is collected.
Is this legal advice?
No. This is a general overview of reported incidents and good practice, not legal or security advice.
Sources
- Malwarebytes: AI chat app leak exposes 300 million messages tied to 25 million users (Feb 9, 2026; Chat & Ask AI, Firebase misconfiguration) (checked 2026-10)
- Biometric Update: Oversecured audit of 17 AI companion apps and the Chattee Chat / GiMe Chat exposure (Mar 19, 2026) (checked 2026-10)
- EDPB: Italian authority fines Replika developer Luka Inc. 5 million euros (decision Apr 10, 2025) (checked 2026-10)
- FTC: inquiry into AI chatbots acting as companions, including use of personal information from conversations (Sept 11, 2025) (checked 2026-10)




