Are AI companions safe to use?

Platform-level safety on the regulated leaders is reasonable. Identity verification on cam sites is now operationally compliant in the UK (Online Safety Act, July 2025) and across five US age-verification states (Texas, Utah, Louisiana, Virginia, Mississippi). The variables under user control are privacy posture (use a non-identifying display name, dedicated email, billing-aware payment), platform choice (operators with named DPOs and EU Representatives score higher on our Privacy and Compliance criteria), and mental-health awareness (treat the persona as software, not a relationship). The full set of criteria and the platform-by-platform scoring live on our scoring page.

What are the absolute red lines for AI companion content?

Four hard-line categories are illegal across all jurisdictions and we refuse to cover any platform that publishes them. (1) Anyone below 18, real or fictional, including AI-generated CSAM, illegal under 18 USC §1466A, the EU CSA Directive 2011/93/EU, UK CJA 2009 §62, Canadian §163.1, and Australian §273.5. (2) Non-consensual content involving real people. (3) Deepfakes of real people without consent. (4) Bestiality. Our publishing checks fail any page that crosses these lines.

What is the UK Online Safety Act?

The UK Online Safety Act 2023 places a statutory duty on services publishing or hosting pornographic content to use highly effective age assurance before granting access. Ofcom is the regulator. Part 5 guidance took effect 25 July 2025. Major cam operators (Chaturbate, Stripchat, LiveJasmin, Jerkmate, BongaCams) ship Ofcom-compliant verification. Major AI companions vary: Candy.ai keeps a self-declaration gate, Joi.ai geo-blocks UK entirely. Maximum fines reach £18 million or 10 percent of global revenue under section 132.

Which US states require age verification?

As of May 2026 the active state age-verification regimes are Texas (HB 1181, 2024), Utah, Louisiana, Virginia, and Mississippi. The Texas Attorney General's $675,000 settlement with Chaturbate in April 2024 was the first major adult-platform enforcement under HB 1181. Chaturbate deployed Incode age verification first for Texas, then extended the same gate to UK traffic for the Online Safety Act. The regulatory map is shifting, so we re-check legal references on every privacy and compliance retest.

Are AI girlfriend apps GDPR compliant?

Compliance varies sharply. Operators with named Data Protection Officers, named EU Representatives, and dedicated policy URLs score highest on our privacy and compliance criteria. Candy.ai (EverAI Limited, Malta C107181) ships the strongest baseline of the AI companions we test. Several brands publish thin privacy policies with no named DPO and vague retention windows, and we flag these explicitly in each review.

Can AI companion use become problematic?

Yes. The variable-reward architecture of synthetic-persona interaction maps onto the same engagement mechanics used in social media and free-to-play mobile gaming, and a subset of users develops compulsive engagement patterns. Our dedicated guide on AI companion addiction covers the published research, the warning signs to watch for, and the mental-health resources we recommend (Crisis Text Line, Samaritans, the international hotline list). We don't promote pathological use patterns, and these safety pages carry no affiliate links.

What happens to my chat data on an AI companion?

Chat content is processed by the platform's language model (frequently a third-party API), stored on the platform's servers, and retained for the period specified in the privacy policy. Major operators name their text, image, voice, and video processors. Smaller operators don't. Our dedicated guide on chat data retention details what we found platform-by-platform. The data-minimization steps under user control are using a non-identifying display name, a dedicated email, and reading the retention window before signup.

How do you score safety in your reviews?

Every platform we cover carries a privacy and compliance score on its scoring page: 14% of the total for AI companions, 16% for cam sites, 14% for adult games, and 14% for real models. That score grades named DPO, named EU Representative, dedicated policy URLs, retention transparency, third-party processor naming, deepfake controls, and age-verification posture. The floor rule is absolute: any platform below 5.0 on privacy and compliance is excluded from our recommendations regardless of commission economics.

Safety & Education

AI Companion Safety 2026: 16 Topics Covered

Read 16 safety guides on AI companions: privacy, addiction, data retention, scams, mental health, UK OSA, US state age laws. Evidence-first, no commercial spin.

By Alexandra JolyLast revised May 2026

By Alexandra Joly, Senior Editor • Last reviewed 2026-05-29 • Methodology • Safety and educational, no affiliate links

Key takeaways

This whole section is safety and educational only. These pages stay strictly clean and carry no affiliate links. The seven questions in the FAQ above each map to a dedicated guide and to the part of our scoring that grades it.
Four absolute red lines we refuse to cover. Anyone below 18 (real or fictional, including AI-generated CSAM). Non-consensual content. Deepfakes of real people. Bestiality. Any page that crosses these lines fails our publishing checks.
The regulatory perimeter spans the UK Online Safety Act (Ofcom Part 5, July 2025, with a further wave in 2026), US state age verification (Texas, Utah, Louisiana, Virginia, Mississippi, where the Texas Attorney General's $675K Chaturbate settlement in April 2024 was the first major enforcement), the EU Digital Services Act, GDPR, and the DMCA.
Privacy and compliance scoring on every platform review grades named DPO, EU Representative, policy URLs, retention, processors, age verification, and deepfake controls. Anything below 5.0 on that score is excluded from our recommendations regardless of commission economics.
The variable-reward architecture of synthetic-persona interaction maps onto the same engagement mechanics as social media and free-to-play gaming. We treat addictive engagement as a real concern, refuse to promote pathological use, and link to non-affiliate mental-health resources (Crisis Text Line, Samaritans, the international hotline list).

This page is the entry point for everything we publish on AI companion safety. Each topic links out to its own guide for the full treatment. Those guides are versioned and re-tested on the cadence laid out on our methodology page, and privacy and compliance scoring gets retested within 7 days of any regulatory news.

The seven safety topics we cover

Are AI companions safe?

The umbrella question. It covers platform-level safety, identity verification, payment-trail privacy, screenshot leakage risks, and the structural variables you can actually control. The privacy and compliance score on every review feeds this question with platform-specific data.

→ Are AI companions safe?

AI companion privacy and data

What chat content is stored, by whom, for how long, and on which infrastructure. It documents named DPOs, named EU Representatives, retention windows, the third-party processors handling text, images, voice, and video, and the data-minimization steps you can take yourself.

→ AI companion privacy and data

AI companion legal landscape

The cross-jurisdictional regulatory regime: the UK Online Safety Act (Ofcom Part 5), US state age verification (Texas, Utah, Louisiana, Virginia, Mississippi), the EU Digital Services Act, GDPR, and the DMCA. It also covers the USC 2257 exemption claims operators make on AI-only content, which remain untested in US federal courts as of 2026 (something we haven't been able to confirm holds up legally).

→ AI companion legal landscape

Protecting minors from AI companion content

The hard line and how it's enforced. Highly effective age assurance (Ofcom-approved methods). The deepfake-of-minor red line (AI-generated CSAM is illegal under multiple statutes regardless of how it was generated). Plus parental controls and reporting procedures.

→ Protecting minors

Do AI girlfriends store my conversation data?

The retention question in detail. Per-platform retention windows, opt-out paths where they exist, deletion request procedures, and what shows up on the credit-card statement (billing descriptor variability).

→ Do AI girlfriends store data?

Are AI girlfriends addictive?

The mental-health concern. Covers the published research on variable-reward engagement mechanics, the warning signs that distinguish casual use from compulsive use, and the non-affiliate mental-health resources we recommend (Crisis Text Line, Samaritans, international hotline list).

→ Are AI girlfriends addictive?

AI companion mental health considerations

The broader mental-health frame. Synthetic-persona use as a complement to, not replacement for, real relationships and professional support. Refusal calibration on edge prompts (consensual-adult only). The honest case for using AI companions, and the honest case against treating them as a substitute for human contact.

→ AI companion mental health

The absolute red lines

Four categories are illegal across every jurisdiction we cover. Any page that crosses one of them fails our publishing checks automatically. We refuse to cover platforms that publish in these categories regardless of commission economics or platform reach.

Red line	Category	Statute coverage
1	Anyone below 18 (real or fictional, including AI-generated CSAM)	18 USC §1466A · EU CSA Directive 2011/93/EU · UK CJA 2009 §62 · Canadian §163.1 · Australian §273.5
2	Non-consensual content involving real people	Cross-jurisdictional revenge-porn and intimate-image abuse statutes
3	Deepfakes of real people without consent	Image-rights, defamation, and emerging deepfake-specific statutes
4	Bestiality	Cross-jurisdictional obscenity and animal-protection statutes

Our editorial process documents how every page is checked against these red lines before it goes live. Our errata log records any corrections we make after publishing.

How we score safety

Most of what people mean by AI companion safety comes down to one score on each platform we cover. Privacy and compliance is a weighted part of every scoring system we publish. The weight changes depending on the type of platform, but the floor rule never moves: any platform scoring below 5.0 on privacy and compliance is excluded from our recommendations regardless of commission economics.

Platform type	Privacy and compliance weight	Re-test cadence
AI companions	14% of the total score	Every 6 months, or within 7 days of regulatory news
Cam sites	16% of the total score	Every 6 months, or within 7 days of regulatory news
Adult games	14% of the total score	Every 6 months, or within 7 days of regulatory news
Real models	14% of the total score	Every 6 months, or within 7 days of regulatory news

The full scoring criteria, the testing protocols, and our policy on flagging anything we haven't tested directly all live on our methodology page. Every score is graded against published criteria rather than a gut feeling, and we lock the scores in place at publish time so an affiliate commission can't quietly talk a 5 into a 7. The mechanics of that lock are documented in our editorial process.

Mental-health resources

These are non-affiliate, non-commercial resources. If you or someone you know is in crisis, contact them directly. We list them on every safety page that touches mental-health considerations.

Crisis Text Line (US, UK, Canada, Ireland): text HOME to 741741 (US), 85258 (UK), 686868 (Canada), or 50808 (Ireland).
Samaritans (UK and Ireland): call 116 123 free, 24/7.
988 Suicide and Crisis Lifeline (US): call or text 988.
International Association for Suicide Prevention: a directory of crisis lines by country at iasp.info/resources/Crisis_Centres/.

Methodology overview: the four scoring systems we apply across every platform we test
About bestgirlfriend.ai and Alexandra Joly: the masthead, the editorial team, and the persona behind the bylines
Editorial process: how we commission, review, and publish each page
Affiliate disclosure: how we lock scores in place and keep affiliate payouts from touching them
Errata: our public correction log
Age verification policy: the site-level age gate plus per-jurisdiction notes