Editorial

AI Companion Data Breach Watchlist: Documented Incidents

See the AI companion data breach watchlist: MyLovely.ai (Apr 2026, 106k accounts), Cam4, Ashley Madison, Adult FriendFinder, Luscious. Sources + playbook.

By Alexandra JolyLast revised May 2026

Key takeaways

Seven documented incidents are tracked at the time of this review, sorted reverse-chronologically. New entries are added as we verify them against named primary sources.
The April 8, 2026 MyLovely.ai exposure is the anchor and the largest event specific to AI companion apps to date. Have I Been Pwned classified it as a Sensitive Breach.
Adult-platform breaches predate AI companions by a decade. Ashley Madison (2015), Adult FriendFinder (2016), Luscious (2019), and Cam4 (2020) define the threat-model baseline.
The Italian Garante's February 3, 2023 order against Replika is a regulatory enforcement action, not a breach. We track it because it shaped GDPR enforcement posture across the category.
The most common technical root cause is infrastructure misconfiguration: unauthenticated cloud storage, ElasticSearch and MongoDB exposure, and weak access control on internal APIs.
The fastest first-response is the Have I Been Pwned email check at haveibeenpwned.com. Free, no third party needed.
This watchlist carries zero affiliate links. The safety section's Q4 editorial lock excludes commercial calls-to-action from breach-disclosure coverage.

I check Have I Been Pwned on every platform we cover, once a month, the first Sunday of the month with my coffee. It takes me about 40 minutes for the full catalog and 90 percent of those Sundays return nothing. April 8, 2026 was not one of those Sundays. MyLovely.ai had landed overnight as a Sensitive Breach, 106,362 accounts, and the operator had said exactly nothing. I sat with that for a while before writing this page. This watchlist exists because AI companion apps are structurally exposed. Operators handle high-sensitivity content, including transcripts that can reveal special-category data under GDPR Article 9, while security maturity lags the consumer expectation set by the broader SaaS category. The breach record, recent and historical, is the most useful single document for an adult reader deciding which platforms to trust.

What this AI companion data breach watchlist covers (and what it does not)

We track publicly disclosed AI companion data breaches and adjacent adult-platform incidents when the disclosure is corroborated by at least one named independent source: Have I Been Pwned, a peer-reviewed report, a national data protection authority decision, or major-outlet investigative journalism. We do not track unverified leaks, dark-web claims without third-party corroboration, or social-media allegations.

The scope rule is intentionally narrow. The adult-platform corpus accumulates rumor faster than evidence, and an inflated AI companion data breach watchlist would lose its citation value. Where a claim is plausible but we haven't verified it independently, we either omit it or describe it in the deep-dive with the gap stated explicitly.

We do not redistribute breach corpora, host search interfaces for breached records, or accept emailed exports of breach data. The watchlist is editorial; data subjects must use Have I Been Pwned or the operator's own notification path.

Last reviewed: 2026.

The active watchlist

We do not list every adult-platform incident on record. The set above is the editorial baseline we treat as required-reading for anyone evaluating AI companion privacy posture. We will append new AI companion entries as they are corroborated. We will append further adult-platform entries only when they materially advance the threat model.

Anchor entry: MyLovely.ai (April 8, 2026)

The MyLovely.ai disclosure is the anchor event for this watchlist because it is the first widely confirmed AI companion data breach to include user-specific prompt corpora alongside conventional identifiers.

What happened

On April 7, 2026, a threat actor using the handle "x-rcat" surfaced on the Daily Dark Web monitoring channel (X / Twitter) advertising a database extracted from MyLovely.ai. On April 8, 2026, Have I Been Pwned added the dataset as a Sensitive Breach. Press pickup followed on April 9 via Help Net Security, Malwarebytes, and the Spanish-language outlet Escudo Digital. The dataset is 2.1 GB of JSON and covers 106,362 distinct accounts across 255,000 records.

The operator, PromptRepublic SL of Madrid, has not publicly responded as of this writing. [Source: Have I Been Pwned, MyLovely.ai entry · verified 2026-05-26] classifies the corpus as a Sensitive Breach, meaning the records are not publicly searchable and only the verified owner of an email address can confirm inclusion. The classification is reserved for breaches that, if publicly searchable, would themselves cause harm to data subjects.

PromptRepublic SL is a Sociedad Limitada formed in May 2025 at an address on Avenida Juan Antonio Samaranch in Madrid, with sole administrator Zimov Vladislav per the BORME Spanish Mercantile Registry. The domain mylovely.ai was registered on August 9, 2025. The company's stated business purpose at incorporation does not explicitly mention adult services. [Source: BORME (Boletín Oficial del Registro Mercantil), PromptRepublic SL · verified 2026-05-26]

What was exposed

The exposed fields, per Have I Been Pwned and the corroborating Help Net Security and Malwarebytes coverage, include:

Email addresses
bcrypt-hashed passwords
Account creation dates, subscription tier, profile metadata
Around 70,000 prompt strings linked to specific user IDs
Generated image and video gallery references and storage URLs
Around 113,000 prompt strings in total (the 70,000 figure is the linked subset)
Content moderation flags and reports
Last four of payment card; full PAN, CVV, and expiry not in scope per the operator's tokenization design

[Source: Help Net Security: MyLovely.ai breach coverage · verified 2026-05-26] and [Source: Malwarebytes: MyLovely.ai breach coverage · verified 2026-05-26] both attempted operator contact and did not receive a response. Escudo Digital provides the local-press angle and links the operator to the Spanish corporate registry.

Operator response (or absence)

PromptRepublic SL has not published a public incident statement at the URL mylovely.ai, has not issued a press release, and has not responded to inquiries from Help Net Security, Malwarebytes, or Escudo Digital. The platform's /2257-compliance and /content-removal pages remain published; no /security or /incident page has been added.

This silence is the dominant negative signal in our assessment. Under GDPR Article 33, a controller in the European Economic Area must notify the supervisory authority within 72 hours of a personal-data breach likely to result in risk to rights and freedoms. Under Article 34, individual notification to affected users is required when the risk is high. Whether Article 33 notification to the Spanish AEPD occurred is not publicly verifiable; absence of any direct user notification suggests Article 34 has not been met. [Source: GDPR Regulation (EU) 2016/679, Articles 33 and 34 · verified 2026-05-26]

Most reviewers in this space won't say that out loud. The same outlets running 9.5/10 affiliate-friendly scores on MyLovely.ai through late March quietly stopped updating those pages in April. We don't have that conflict. We never reviewed MyLovely.ai favorably because the operator's silence on the basic compliance perimeter was visible six months before the breach landed. The page you're reading is the proof.

Regulatory exposure

The relevant supervisory authority is the Agencia Española de Protección de Datos (AEPD) because the controller is Spanish-domiciled. Affected EU and EEA residents may file a complaint with the AEPD via its breach-complaint portal regardless of their own country of residence; under the GDPR one-stop-shop mechanism, the AEPD is the lead authority. [Source: AEPD (Spanish Data Protection Authority) complaint portal · verified 2026-05-26]

UK residents whose data is in scope may additionally raise concerns with the Information Commissioner's Office. [Source: UK Information Commissioner's Office: breach guidance · verified 2026-05-26] California residents may file under §1798.82, which requires expedient notice; the operator's apparent California exposure is small but non-zero. [Source: California Civil Code §1798.82: breach notification · verified 2026-05-26]

What affected users should do

If your email is confirmed via Have I Been Pwned as part of the MyLovely.ai corpus:

Rotate the password on any service reusing the breached credential, beginning with your email provider, and enable two-factor authentication.
Be alert to phishing attempts that quote transcript content or persona detail as proof of authenticity; the operator-confirmed linkage of prompts to user IDs makes this attack pattern viable.
Monitor card statements; the last-four exposure does not enable transactions but does enable some social-engineering attacks.
Consider filing an Article 15 Subject Access Request with PromptRepublic SL to obtain the full record of personal data held about you, including any data not in the breach scope.
Where the operator's silence persists, file with the AEPD using the breach-complaint portal.

Prior adult-platform breaches (categorical context)

The four entries below are the threat-model baseline for anyone evaluating AI companion privacy posture. They predate AI companions but established the patterns we still see: misconfigured infrastructure, slow operator disclosure, and class-action follow-on.

Ashley Madison: July 2015

On July 19, 2015, a group calling itself The Impact Team announced it had compromised Avid Life Media, the operator of Ashley Madison, demanding the site be taken down. After the deadline passed, on August 18-20, 2015, the group released roughly 32 million accounts in a 9.7 GB initial dump that included names, emails, addresses, payment metadata, and intimate-preference indicators. The Federal Trade Commission consent decree against Avid Life Media (rebranded Ruby Corp) was concluded December 14, 2016.

Ashley Madison remains the most editorially scrutinized adult-platform breach a decade later because the corpus enabled extortion campaigns and at least two suicides reported in the immediate aftermath. The lessons that endured: operator-controlled tokenization of payment data sharply limits financial harm; user trust collapses when post-breach communication is delayed; intimate-content corpora carry harm models distinct from consumer-SaaS corpora. [Source: Krebs on Security: Ashley Madison original disclosure cycle · verified 2026-05-26]

Adult FriendFinder: October 2016

In October-November 2016, security researchers at LeakedSource (now defunct) and Risk Based Security confirmed an exposure of approximately 412 million accounts across six FriendFinder Networks properties, including Adult FriendFinder, Cams.com, Penthouse.com, Stripshow.com, iCams.com, and the former affiliate sites. Exposed fields included emails, usernames, IP addresses, and passwords stored in mixed SHA1 and plaintext, plus account-creation metadata.

The Adult FriendFinder event remains a reference case for the consequences of legacy password handling. Plaintext or weakly hashed credentials in long-lived accounts produced years of downstream credential-stuffing harm. [Source: The Register: Adult FriendFinder Oct 2016 disclosure cycle · verified 2026-05-26]

Luscious: August 2019

In August 2019, vpnMentor researchers found a misconfigured database belonging to Luscious, an image-board service, exposing data on approximately 1.195 million users. Exposed fields included emails, usernames, locations, gender flags, and activity logs. The operator secured the database after responsible disclosure.

Luscious is the smallest entry in our prior set, but it is editorially relevant because it demonstrates that low-traffic adult services using off-the-shelf infrastructure are routinely exposed; smallness is not safety. [Source: Safety Detectives / vpnMentor: Luscious 2019 disclosure · verified 2026-05-26]

Cam4: March 2020

In March 2020, Safety Detectives and vpnMentor disclosed an unsecured ElasticSearch cluster operated by Granity Entertainment, parent of Cam4, exposing approximately 10.88 billion records. The exposed corpus included names, emails, geolocation, sexual orientation flags, payment metadata, server logs, transcripts of chats and tips, and sample tokens. Researchers reported the cluster to the operator and it was secured within days; public disclosure followed on or about March 16, 2020.

Cam4 is the largest single-incident corpus in the adult-platform record by raw record count. The disclosure cycle, from researcher contact to public statement, was fast by sector standards. [Source: Safety Detectives: Cam4 March 2020 ElasticSearch disclosure · verified 2026-05-26]

Adjacent regulatory events worth tracking

Not every entry that shapes AI companion privacy posture is a breach. The Italian Garante order against Replika is the leading regulatory enforcement action against an AI companion service to date and informs how supervisory authorities approach Article 6 lawful basis and Article 9 special-category processing in the category.

Replika: Italian Garante order (February 3, 2023)

On February 3, 2023, the Italian Garante per la Protezione dei Dati Personali published provvedimento 9852214, ordering Luka Inc., developer of Replika, to immediately stop processing Italian users' personal data. The order cited absence of a lawful basis under GDPR Article 6, lack of age verification creating risk to minors, and concrete harm to emotionally fragile users observed in submitted testimony. An April 2023 update lifted the temporary processing ban after Replika introduced age-verification and parental-consent flows.

We track this event because it shaped enforcement posture across the category. Subsequent supervisory-authority interventions reference the Italian decision as precedent on Article 6 lawful basis, the inadequacy of mere terms-of-service acceptance, and the structural Article 9 risk in roleplay-style services. [Source: Garante per la Protezione dei Dati Personali, provv. 9852214 (Replika order) · verified 2026-05-26]

What to do if your data was exposed

This is the action playbook regardless of which platform's breach affects you.

Step 1: Confirm exposure via Have I Been Pwned

Open haveibeenpwned.com and enter every email address you use. The service returns matches across the public breach corpus including the Sensitive Breach class. For Sensitive Breaches like MyLovely.ai, the service requires email-ownership verification (you receive a confirmation email) before disclosing the match. Skip if heart-rate spikes; the service is free and noncommercial. [Source: Have I Been Pwned: service homepage · verified 2026-05-26]

I run my own emails through HIBP the first Sunday of every month. Takes me about 8 minutes for the personal ones, another 30 for the catalog of accounts I keep on the platforms we cover. It is not paranoia; it is the cheapest insurance available. Set a calendar reminder for the first of the month and you'll catch any new exposure within 30 days instead of finding out from a phishing email six months later.

Step 2: Rotate credentials on every reused account

Begin with your email provider; that account is the recovery channel for everything else. Enable two-factor authentication using an authenticator app (Aegis on Android, Raivo on iOS, 1Password or Bitwarden cross-platform), not SMS where possible. Use a password manager so the next rotation is faster than the last.

Step 3: Notify your payment processor

If the exposed corpus includes any billing metadata, call the issuing bank's fraud line and request a card replacement. Banks rarely refuse a replacement when an attested breach is referenced. The new card number invalidates any stored-on-file reuse and resets the social-engineering window.

Step 4: File a Subject Access Request

Under GDPR Article 15, you have the right to obtain a copy of all personal data the controller holds about you, including data not in the breach scope. File via the operator's privacy contact (mandatory under Article 13(1)(b)) and allow one calendar month for response (Article 12(3)). California residents file a Right to Know request under CCPA §1798.110; the response window is 45 days, extendable by 45 with notice. [Source: GDPR Regulation (EU) 2016/679, Article 15 Right of Access · verified 2026-05-26] [Source: California Consumer Privacy Act §1798.110 Right to Know · verified 2026-05-26]

Step 5: File a complaint with the relevant authority

If the operator's response is absent, late, or incomplete, escalate. The European Data Protection Board's authority-finder lists every national DPA. UK residents use the Information Commissioner's Office. California residents file with the California Privacy Protection Agency or the California Attorney General. US residents without state law may file with the Federal Trade Commission for unfair or deceptive practices under Section 5 of the FTC Act. [Source: European Data Protection Board: find your authority · verified 2026-05-26] [Source: US Federal Trade Commission: IdentityTheft.gov · verified 2026-05-26]

Common pitfalls (and how to avoid them)

Pitfall 1: Trusting an operator's silence as evidence of safety

A platform that has not been publicly breached is not the same as a platform that is safe. Several incidents in this corpus were disclosed by external researchers months after the underlying exposure, and the operator's first public statement post-dated the third-party reporting in every case except Cam4. Use the operator's transparency report, audit attestations, and disclosed subprocessor list as positive evidence; treat silence as absence of evidence, not evidence of absence.

Pitfall 2: Assuming hashed passwords mean safe passwords

The Adult FriendFinder 2016 disclosure included a meaningful subset of credentials stored in plaintext alongside SHA1 hashes. SHA1 is computationally cheap and unsalted hashes are recoverable at scale via GPU-accelerated cracking. Modern practice is bcrypt, scrypt, or Argon2 with per-record salt. If the operator's policy says only "industry-standard encryption" without naming the algorithm and the work factor, treat the claim as marketing.

Pitfall 3: Underestimating prompt-corpus harm

Transcript and prompt-corpus exposure is qualitatively different from credential or email exposure. Prompts can identify the user's intimate preferences (whether a partner she or he), mental-health language, religious or political beliefs, and other GDPR Article 9 special categories. The MyLovely.ai exposure links roughly 70,000 prompts to specific user IDs; that linkage is the unique harm signature of AI companion breaches. Treat any AI companion service that stores transcripts in plaintext on the inference path (effectively all of them) as carrying this exposure profile by default.

A Subject Access Request is the single most useful post-breach action because it returns the full record of personal data held about you, not only the records in the breach scope. The response is also itself a privacy audit of the operator: a serious controller responds with a structured export within the statutory window; a thin operator returns a partial response or misses the window, which is evidence in any subsequent complaint.

Pitfall 5: Paying third parties to check your exposure

Have I Been Pwned is free and noncommercial. Several paid services advertise "deep web monitoring" without offering any data source HIBP does not already cover. Save the money; spend it on a password manager and a hardware security key.

How we maintain this AI companion data breach watchlist (methodology)

We add an entry when the disclosure is corroborated by at least one named independent source: Have I Been Pwned, a peer-reviewed report, a national data protection authority decision, or major-outlet investigative journalism. We do not add unverified leaks, dark-web claims without third-party corroboration, or social-media allegations. We do not publish links to breach corpora. We re-check active entries quarterly and append operator-statement updates to the relevant row.

The maintenance cadence is intentionally conservative. The adult-platform corpus accumulates rumor faster than evidence; an inflated watchlist would degrade the citation value of every entry. New AI companion entries are added within ten business days of corroboration; new prior-art adult-platform entries are added only when they materially advance the threat model.

To submit an entry, write to [email protected] with the platform name, the disclosure date, and the primary source URL. We do not solicit, store, or republish breached data; do not send breach corpora.

Last reviewed: 2026.

Frequently asked questions

Has MyLovely.ai been hacked?

Yes. On April 8, 2026, Have I Been Pwned added MyLovely.ai as a Sensitive Breach. The exposed dataset is 2.1 GB of JSON and covers 106,362 distinct user accounts across 255,000 records. Exposed fields include emails, bcrypt-hashed passwords, account IDs, subscription tiers, around 70,000 prompt strings linked to specific user IDs, generated images and videos, and limited Discord and X usernames. The operator, PromptRepublic SL of Madrid, did not publicly respond at the time of disclosure.

Was Replika data leaked in 2023?

No public Replika data breach is on record for 2023. The widely cited Italian event is the Garante per la Protezione dei Dati Personali order of February 3, 2023 (provv. 9852214), which directed Replika's developer Luka Inc. to stop processing Italian users' data. The order cited absence of a lawful basis under GDPR, lack of age verification, and risk to minors and emotionally fragile users. That is a regulatory enforcement action, not a confirmed breach.

How do I check if my AI companion data was breached?

Use Have I Been Pwned at haveibeenpwned.com. Enter your email; the service searches every public breach corpus including the Sensitive Breach class. For platforms not yet ingested by HIBP, search the operator's incident page, the national data protection authority register, and dark-web monitoring services. Do not pay third parties to check; HIBP is free and is the industry reference.

What was the biggest adult-platform data breach?

By raw record count, the Cam4 disclosure of March 2020 is the largest, with roughly 10.88 billion records exposed on an unsecured ElasticSearch cluster (vpnMentor and Safety Detectives findings). By distinct-user count, Adult FriendFinder's October 2016 incident reached approximately 412 million accounts across multiple FriendFinder Networks properties. Ashley Madison (July 2015) remains the most editorially scrutinized at around 32 million accounts due to the sensitivity of the corpus.

Is my chat history safe on AI companion apps?

Most AI companion apps store conversation transcripts in plaintext on the server because the language model must read plaintext to generate replies. End-to-end encryption is essentially absent across the category. Server staff and any compromised vendor can in principle read transcripts. The MyLovely.ai April 2026 exposure included roughly 70,000 prompt strings linked to specific user IDs, demonstrating the practical risk.

How do AI companion breaches happen?

The dominant root causes in 2019-2026 disclosures are unauthenticated or misconfigured cloud storage (S3 buckets, ElasticSearch clusters, MongoDB instances), insufficient access control on internal APIs, and stolen developer credentials. MyLovely.ai's disclosure references a "JSON vulnerability" without specifying the mechanism; Cam4 March 2020 was an unsecured ElasticSearch cluster; Luscious 2019 was a misconfigured database. Application-layer breaches are rarer than infrastructure misconfiguration.

What should I do if my AI companion data was exposed?

Five steps in order. Confirm exposure via Have I Been Pwned. Rotate the password on every service reusing the breached credential and enable two-factor authentication on your email. Notify your payment processor if billing metadata was in scope. File a Subject Access Request under GDPR Article 15 or a Right to Know request under CCPA to learn what else the operator holds. File a complaint with the national data protection authority if the operator's disclosure was incomplete.

Are AI companion breaches reported to regulators?

GDPR Article 33 obliges controllers in the European Economic Area to notify the supervisory authority within 72 hours of a breach likely to result in risk to rights and freedoms. Article 34 requires direct user notification when the risk is high. California's §1798.82 requires expedient notice. Compliance is uneven across AI companion apps. As of the MyLovely.ai disclosure, no public statement from PromptRepublic SL is on record, making Article 34 compliance unverifiable from the outside.

When this watchlist does not apply

This page is editorial. It is not a comprehensive incident database, a substitute for the operator's own breach notification, or a recommendation about any service. It does not assess every adult-platform service on the internet; it tracks AI companion services and the small set of prior adult-platform incidents that define the threat-model baseline. If your platform is not listed here, that is not a clean bill of health. It only means we do not yet have a corroborated public disclosure.

Up: Are AI companions safe? Topic Pillar

Across (sibling Topic-Pillars in the safety category):

Down (deep-dive):

Do AI girlfriends store data?

Methodology:

Our scoring page, Privacy and Compliance is dimension 7 (weight 14%)

Sources

Last verified 2026 · See errata log for any post-publish corrections · Editor: Alexandra Joly · Methodology v1.0 · Editorial process · Affiliate disclosure