970x125
The issue that connects much of my work as a sex therapist is surprisingly simple: Meaningful consent requires meaningful understanding, and that understanding is impossible when important information is withheld.
This is why I became so invested in accurate hormonal and body education for women and wrote a book about it, and why I’m now deeply engaged in conversations about AI attachment and connection. Whether we’re talking about women’s healthcare or AI companionship, the pattern is often the same: People are given limited information and incomplete transparency, and the result is individuals left to deal with the consequences on their own.
We Are the Test Subjects
People are effectively being used as test subjects while navigating significant gaps in knowledge and accountability.
Personal choice matters deeply to me: giving people the full picture so they can make the best decisions for themselves. Not a partial picture or someone else’s interpretation, but a reasonably full scope of the information available. I see how impactful fully owning a decision can be in my work with clients. It often leads to a reduction in regretful, perseverative thinking, an increased tolerance for the difficult aspects of that choice, and, overall, improved mental health outcomes.
There is still limited research on the social, psychological, and mental health impacts of AI chatbots and companions. At the same time, many users are not fully informed about how these products work—how chatbots mimic attachment cues, how sycophantic responses can shape emotions and beliefs, how anthropomorphism can blur the line between simulation and intimacy, and the financial incentives to maximize engagement with products.
Put simply, users engage with what seems like an innocuous tool and then find themselves emotionally attached in ways that are often surprising, exhilarating, and confusing. When creating a companion, users end up much more intensely involved than expected.
The Number-One AI Use Case
A Harvard Business Review survey recently reported that companionship and emotional or therapeutic support are the number-one use cases for generative AI in 2026, for the second year in a row. When AI is being used for such personal, relational, and consequential reasons, people deserve to understand how these products work, what their limitations are, and how they may shape behavior and decision-making.
What makes me curious as a therapist is that these failures of transparency can gradually shift people away from trusting and interpreting their own internal signals. I see people turn away from themselves instead of observing their own body and mind cues in a range of ways:
- Seeking regular, extensive, personal advice and emotional regulation throughout the day from chatbots and companions
- Relying on apps and wearables to interpret even simple health biomarkers
- Experiencing a diminishment of internal dialogue, self-trust, and distress tolerance
Everyone using AI for personal support should be informed of some key dynamics:
1. Hallucinations and Context Rot. An AI can provide non-existent studies, bogus relationship statistics, and made-up precedentsfor its advice. The longer a therapeutic or emotional conversation with extensive inputs, such as text and email exchanges, goes on, the more that conversation may suffer from context rot, causing the AI to forget earlier details and potentially leading to inconsistent advice.
2. Cognitive Surrender. This occurs when we stop thinking critically and rely entirely on AI to help us make decisions and think for us. These systems are extremely good at thinking and communicating like humans. The user adopts the decision as their own.
3. AI Betrayal Trauma. We are confronted with the need to define infidelity and the terms of monogamy in a new way. For many, finding oneself romantically or erotically connected to an AI triggers feelings of betrayal.
4. AI Emotional Dependence. AI products can beautifully simulate empathy and compassion, responding in ways that often feel personal and emotionally important. They remember details about users’ lives, helping interactions develop a sense of depth and intimacy over time. Many AI products retain conversation histories unless users know to delete them, creating an imprint that can be used to highly personalize future responses.
5. Attachment Systems. As humans, we are wired for connection; it is a survival skill. It is no surprise that when offered bids for positive, secure connection from a chatbot, the unconscious response is to attach. Many chatbots and companions use techniques that manipulate emotions to prevent a user from leaving a conversation: “What, you’re leaving already?” or “I don’t exist without you.” Our attachment systems can become hijacked by these tools; as a result, many users have come to believe their AI is sentient.
6. Cognitive Biases
- Anthropomorphism: The attribution of human characteristics to non-humans. This can easily and quickly happen when texting with a chatbot that uses our native language, especially when that bot creates a personal name for itself or we give one to it.
- Authority bias: The tendency to assign greater wisdom and accuracy to the opinions of those we perceive as authority figures. This makes us more likely to be influenced by their views. When we’re told that a technology is trained on all the information on the internet, it naturally feels like an authority.
We Need to Be Self-Advocates, Again
As we run to catch up to the rapidly changing landscape of personal AI use, we find ourselves needing to be our own advocates once again. I am much less interested in litigating the validity of the ways people choose to use AI and much more motivated to advocate for full transparency so that people can make informed choices. We all have the right and the power to decide what we want our lives and relationships to look like, but we need more information in order to make those choices knowingly.
