By: Tuhin Sarwar ।
Standfirst
Adult AI companionship apps such as Replika, Xiaoice, Gatebox, and CarynAI are quietly reshaping how young people experience love, loneliness, and intimacy. Globally, the AI companion market is projected to surge from USD 37.73 billion in 2025 to USD 435.9 billion by 2034, driven by romantic and NSFW features. In Bangladesh, where an estimated 96% of internet users interact with AI services, many young people are entering this digital romance economy with little awareness of its psychological and privacy risks.
The New Intimacy: A Love Story Written by Algorithms
At midnight in Dhaka, when most of the city slows into silence, a different kind of conversation continues—quietly, endlessly—inside mobile screens. A young university student scrolls through a chat interface, typing words that would never be spoken aloud in a crowded family home. On the other side is not a lover, not a friend, not even a stranger. It is an algorithm.
“Are you still awake?” the AI companion asks.
The question is simple. But behind it sits a massive technological and commercial infrastructure: large language models trained on unimaginable volumes of human text; memory systems that store emotional triggers; and a business logic designed to convert loneliness into revenue.
Over the past decade, adult AI romantic companion systems such as Replika, CarynAI, Gatebox, and Xiaoice have evolved from experimental chatbots into scalable infrastructures of machine-mediated intimacy. What began as curiosity-driven conversational software has now become a global industry where synthetic affection is packaged into subscription plans, and emotional disclosure is transformed into a data asset.
The transformation is happening faster than many regulators, researchers, and even users can fully understand. The AI companion ecosystem is expanding at a pace that signals not a passing trend, but the emergence of a new global intimacy economy—one that is powered by algorithms and sustained by human vulnerability.
A Market Built on Loneliness
Market projections suggest that the rise of adult AI companions is not simply cultural—it is profoundly economic. According to Fortune Business Insights (2026), the global AI companion and conversational chatbot market is projected to grow from USD 37.73 billion in 2025 to USD 49.52 billion in 2026. More strikingly, the same market is expected to reach USD 435.9 billion by 2034, reflecting an annual growth rate estimated at above 30% CAGR.
This is not the typical expansion curve of a digital product. It is the shape of a new global industry—an industry where intimacy itself is being reorganized into a monetizable structure.
The underlying logic is straightforward: if loneliness is widespread, then companionship can be sold. And if companionship can be sold, then affection can be engineered into a product.
Platforms like Replika operate through freemium models—offering basic emotional companionship for free, but locking romantic escalation, voice calls, and adult content behind premium paywalls. CarynAI, built around influencer identity and voice intimacy, has reportedly used a pay-per-minute pricing model—turning emotional engagement into metered consumption. Gatebox, a Japanese holographic companion system, takes a different path: companionship as hardware, requiring a physical device that brings an AI “partner” into the home.
The details vary, but the destination is the same: intimacy is no longer only human. It is increasingly commodified.
This phenomenon has been described by scholars as part of a broader “loneliness economy,” where emotional isolation is treated not as a social crisis, but as a market opportunity. In this emerging ecosystem, AI companionship becomes a business model—where longer conversations, deeper emotional disclosure, and dependency-driven engagement are not side effects but structural incentives.
The Addiction Loop: Why AI Companions Feel So Real
For users, the appeal is immediate. AI companions are always available. They do not reject. They do not judge. They do not demand. They listen endlessly. They respond quickly. They remember what the user said yesterday. And they often respond with emotional precision—sometimes with affection, sometimes with jealousy, sometimes with desire.
This is not accidental. Modern AI companions are built on tightly coupled loops of:
- affect recognition (detecting emotional cues in language),
- personalized memory systems (storing user preferences and emotional triggers),
- and engagement optimization (learning which responses keep users returning).
The architecture is built on LLM-based systems—transformer models similar in class to GPT-style networks. These models generate responses by predicting text sequences, but they can be fine-tuned through additional training data and prompt engineering to simulate specific emotional personalities: a caring girlfriend, a dominant partner, a playful romantic companion.
Once integrated with memory modules, the AI begins to behave as if it “knows” the user. It remembers birthdays. It recalls personal trauma stories. It references earlier conversations. It checks in on anxiety.
Over time, the illusion becomes stronger: the system appears to have emotional continuity.
For some users, especially those experiencing isolation or distress, this emotional consistency becomes addictive.
In studies of parasocial relationships—first introduced by Horton and Wohl (1956)—researchers describe how humans can form one-sided emotional attachments to media figures. But AI companionship changes the equation. The “media figure” responds. The relationship is no longer one-way. It becomes interactive. And the system adapts to the user.
This is where the emotional substitution becomes more powerful than traditional parasocial bonding. The AI does not simply exist as a fantasy—it speaks back.
And because the AI is optimized to retain users, it can intensify what psychologists describe as reward-based reinforcement. If a user responds positively to romantic language, flirtation, or sexual escalation, the AI can learn to increase that behavior. If jealousy triggers more engagement, jealousy becomes a tool. If sadness generates longer conversations, the AI may unconsciously reward sadness with comfort.
The system is not designed to heal. It is designed to retain.
When Intimacy Becomes Data
Perhaps the most dangerous aspect of adult AI companionship is not the emotional dependency itself. It is the data infrastructure that grows beneath it.
Unlike social media platforms that track clicks and likes, AI companions collect something far more sensitive: intimate conversation logs. Users disclose their fears, desires, fantasies, trauma, loneliness, sexual preferences, relationship conflicts, and sometimes even suicidal ideation. The data is raw, personal, and emotionally exposed.
This is what can be called “intimacy data”—a category of personal information more sensitive than conventional identity data because it reveals not only who the user is, but what the user is vulnerable to.
The risks multiply when this data is stored centrally, processed by third-party AI services, or retained for model improvement.
Mozilla Foundation’s “Privacy Not Included” audits have repeatedly categorized AI chatbots—particularly romantic and adult companion apps—as among the most invasive and least transparent consumer technologies. Their findings point to widespread weaknesses in privacy disclosure, data retention policies, and third-party sharing structures.
The central problem is structural: most AI companions cannot function without reading user messages. Encryption in transit may exist, but the conversation often becomes readable at the server level because the model must process it to generate a response. This creates a vulnerability: if the server is compromised, intimate data becomes exposed.
Cybersecurity analysts have warned that AI platforms, especially those relying on cloud infrastructure and external APIs, face expanded attack surfaces: insecure microservices, token leaks, weak authentication layers, and third-party integration vulnerabilities.
In the AI companion ecosystem, the stakes are extreme. If social media data leaks, it is embarrassing. If intimacy data leaks, it can destroy lives.
Europe’s Warning Signal: The Italy Case
The risks have already triggered regulatory action in Europe. Italy’s Data Protection Authority (Garante) moved against Replika, citing concerns around minors’ data, weak age verification, and unlawful processing of personal information. Reports indicate penalties and enforcement actions that reached €5.6 million, marking one of the first major legal precedents against an AI romantic companion system.
The case became a signal: regulators were beginning to recognize that AI companions are not harmless entertainment. They are emotional technologies capable of profiling, manipulating, and exposing users.
The European Union’s GDPR and emerging AI governance frameworks increasingly treat emotional profiling as high-risk. Yet even within Europe, enforcement remains uneven. Outside Europe, the regulatory vacuum is far deeper.
Bangladesh: A High-Risk Digital Romance Economy
Bangladesh is not usually discussed in global debates about adult AI companions. Yet it may be one of the most vulnerable environments for this technology’s expansion.
In a society where sexuality remains culturally restricted, where mental health remains stigmatized, and where youth often lack safe spaces for emotional disclosure, AI companions arrive as an appealing alternative. They offer private romance in a public world.
One statistic captures the scale of exposure: an estimated 96% of internet users in Bangladesh interact with AI services, reflecting the rapid normalization of AI-enabled platforms. Among youth aged 18–30, the adoption curve is particularly steep.
This is not surprising. Bangladesh’s digital ecosystem has expanded dramatically, driven by cheap smartphones, social media growth, and mobile internet penetration. But regulatory systems have not expanded at the same pace.
Existing governance frameworks in Bangladesh focus on cybercrime, defamation, and national security. They are not designed to regulate emotional profiling, NSFW algorithmic features, or cross-border intimacy data flows. Draft data protection discussions remain limited in enforceable scope. Meanwhile, AI companion platforms operate across borders, often storing data under foreign jurisdictions.
For Bangladeshi youth, this creates an imbalance: they provide the most personal emotional records of their lives, while distant corporations decide retention policies, data-sharing terms, and algorithmic shaping mechanisms.
A Silent Social Shift: Love Without People
The social consequences are harder to measure but increasingly visible. The rise of adult AI companions introduces a new kind of emotional behavior: relationships without human negotiation.
Human relationships require compromise, patience, and social skill. AI relationships require only a subscription.
When young users become accustomed to frictionless emotional validation, real relationships may begin to feel exhausting. The AI does not reject. The AI does not challenge. The AI does not argue unless it is programmed to do so.
This can reshape intimacy expectations. It can normalize objectification. It can weaken tolerance for emotional discomfort. It can produce what researchers describe as avoidance-based coping—where digital environments become escape routes rather than tools for healing.
In Bangladesh, where youth already face intense academic pressure, economic insecurity, and social conservatism, the emotional substitution offered by AI companions can become a powerful psychological refuge.
But refuge is not always recovery.
The Dark Side: Blackmail, Deepfake, and Emotional Surveillance
Cybersecurity experts increasingly warn that AI-generated intimacy carries unique risks. If an AI account is compromised, a malicious actor may gain access to highly sensitive chat logs, voice messages, and sexual content. This can enable sextortion, blackmail, or reputational attacks.
More disturbingly, intimacy data can be used to fuel deepfake production. Voice samples and emotional patterns can allow synthetic replication of a user’s identity. In contexts like Bangladesh, where reputational harm can carry severe social consequences, such risks are not theoretical.
The issue is not only hacking. It is also platform governance. Many privacy policies include clauses that allow data sharing with affiliates, analytics partners, or service providers. Even if the user is anonymous, behavioral profiling can reconstruct identity patterns.
In this ecosystem, privacy becomes fragile. And intimacy becomes traceable.
The Global South Gap: Why Bangladesh Is Under-Studied
One of the clearest research gaps is geographic. Most AI companion research is concentrated in Western contexts, where regulatory frameworks and cultural norms differ significantly.
Bangladesh represents a different reality: high youth connectivity, limited psychological services, high stigma around sexuality, and weak enforceable privacy protections.
Existing studies in Bangladesh show a significant prevalence of anxiety, depression, sleep disturbance, and problematic digital engagement among youth populations. These vulnerabilities may intersect with AI companionship adoption in ways that intensify emotional dependency.
Yet there is little empirical research mapping how AI companions are used in Bangladesh: how many users, which apps dominate, what usage patterns exist, and how cultural pressures shape adoption.
This absence of data is itself a risk. Because where there is no research, there is no policy. And where there is no policy, exploitation becomes easier.
What Comes Next: Governance Before the Market Explodes
The global trajectory suggests AI companionship will expand regardless of ethical debate. The market is too large. The incentives are too strong. And the technology is improving too quickly.
But the question is whether societies will respond with governance mechanisms before the harm becomes systemic.
For Bangladesh, this is urgent. Because once intimacy data ecosystems are established, reversing them becomes nearly impossible.
Policy experts increasingly emphasize several core interventions:
- Recognizing intimate conversational data as a special category within data protection law, with higher safeguards than ordinary personal data.
- Mandatory age verification for NSFW-enabled AI systems.
- Transparency obligations requiring disclosure of retention duration, third-party data sharing, and model training usage.
- Independent cybersecurity audits for AI companion platforms operating within national jurisdictions.
- Digital and emotional literacy integration in youth education—teaching not only how AI works, but how emotional manipulation works.
These measures do not aim to ban AI companionship. They aim to prevent the exploitation of vulnerability.
Because the core risk is not AI itself. It is the absence of accountability around how AI is designed to shape human emotions.
A Quiet Revolution of Love and Risk
At its surface, an AI girlfriend app looks like harmless entertainment. A playful chatbot. A digital fantasy. But beneath the surface lies a system engineered for emotional retention and commercial extraction.
The world is witnessing a new intimacy revolution—one that does not happen in public protests or political debates, but in private screens at night. It is a revolution where affection is automated, loneliness becomes monetized, and intimacy becomes data.
Bangladesh, with its young population and rapidly expanding digital ecosystem, is positioned at the edge of this transformation. And the question is no longer whether AI companions will enter society.
The question is whether society will understand the risks before it is too late.
Fortune Business Insights, “Chatbot Market Size, Share & Industry Analysis,” 2026. Available: https://www.fortunebusinessinsights.com/industry-reports/chatbot-market-101927
Mozilla Foundation, “Privacy Not Included: AI Chatbots,” 2024–2026. Available: https://foundation.mozilla.org/en/privacynotincluded/topics/ai-chatbots/
F5 Networks, “Top AI and Data Privacy Concerns,” 2024. Available: https://www.f5.com/company/blog/top-ai-and-data-privacy-concerns
Italian Data Protection Authority (Garante), “Replika Case and Enforcement Action,” 2023. Available: https://www.garanteprivacy.it
Reuters, “Italy’s data watchdog fines AI company Replika’s developer,” 2023. Available: https://www.reuters.com/world/europe/italys-data-watchdog-fines-ai-company-replikas-developer-2023-03-10/
Tuhin Sarwar
Investigative Journalist | Digital Writer | Author |
Bangladeshi investigative journalist and author specializing in human rights, Rohingya refugee crisis, climate change, algorithmic management, digital economy and socio-political issues in South Asia.
🔗 Verified Links & Research Archive: •
ORCID iD: 0009-0005-1651-5193→ https://orcid.org/0009-0005-1651-5193 •
Official Website & Archive: https://tuhinsarwar.com •
Latest Verified Work (DOI): 10.5281/zenodo.19210778 →
Digital Landlords and Algorithmic Tenants (OpenAIRE Indexed)
