AI companion safety & privacy
These apps are built to feel intimate, which is exactly why the privacy and safety questions are more serious than for an ordinary chatbot. This is the evidence layer: a maintained breach tracker, a sourced privacy index, and clear guidance — so you can judge an app on what it actually does with the most personal data you will ever type.
AI Companion Data-Breach Tracker
A maintained, dated log of every documented breach, exposure and regulatory finding in the category — what was exposed, how serious it was, and the original source for each.
Open the tracker → Data assetAI Companion Privacy Risk Index
A sourced comparison of 16 apps on the things that matter: whether they train on your chats, encryption, data sale, deletion and jurisdiction.
Compare all apps → GuideWhat AI companions collect — and how to check
A plain-English guide to the data these apps gather, the policy red flags to look for, and how to use one with a smaller footprint.
Read the guide → GuideSafety: minors, risks and safer use
Age assurance, vulnerable users and the harms regulators and lawsuits are now focused on — and what safer adult use looks like.
Read the guide → GuideHow to spot an AI companion scam
Twelve warning signs — fake domains, hidden renewals, in-character money requests and impersonation — plus a five-minute safe-companion check.
Read the guide → GuideAI companion laws in 2026
What the UK, EU, US, China and Australia now require — age checks, AI disclosure, intimate-image offences and the rules landing next.
Read the guide → GuideWhy AI companion apps ask for ID
Age checks explained — what each method collects, who processes your selfie or ID, how long it is kept, and how to spot a fake verification screen.
Read the guide → RankingBest AI companion apps for privacy
The apps that handle your data most responsibly, ranked from the same scoring method as every other grade on the site.
See the ranking → MethodHow we test — and why privacy is 20% of the grade
Our fixed scoring method weights privacy at a fifth of every headline grade. Here is exactly how we assess it.
See the method →Everything here is research-assessed from public reporting, regulatory records and each provider’s own policy — clearly dated, sourced and conservative. We say “not disclosed” when a provider is silent rather than assuming the worst, and we flag genuine contradictions as “contested”. It is not a security audit or legal advice. Found an error? info@companionscore.com.