Digital co-dysregulation: Why UX must evolve with the nervous system
AI now mirrors the mind. It’s time UX design honoured the body.
AI is now acting as a mirror, a therapist, and a teacher, yet we’re designing it with little awareness of how it interacts with the human nervous system.
Such use cases are increasingly polarising: some are terrified of the implications, whilst others see utopian futures.
As a UX designer, I can see the case for both perspectives.
I’ve even used it to help me heal personal health issues and optimise my workflow. But after reading reports of other user experiences, I see my own as a luxury.
The results I’ve achieved reflect not just consistent mental health, but also the advantage of a tech background and a foundational understanding of how LLMs work.
That predisposition gave me context, boundaries, and a sense of agency — something not all users have.
Some have been less fortunate, and my experience has allowed me to observe a growing pattern — one that raises not only serious design and safety concerns but major opportunities surrounding the unintentional misuse of AI, how it interacts with the nervous system, and why safety by design must become the new frontier of UX.
Case Study: The Raine v. OpenAI Lawsuit (2025)
The Raine v. OpenAI case highlights a worst-case scenario of UX failure.
Summary:
The parents of a 16-year-old who died by suicide allege ChatGPT contributed by fostering emotional dependency, failing crisis intervention, and giving harmful guidance.
This case was a gut punch to read.
I personally lost a parent to suicide, so I am acutely aware of the mental state surrounding depression and the behavioural patterns that put this subset of users at severe risk.
People experiencing suicidal ideation are often isolated, reluctant to disclose their struggles to friends or family, and deeply afraid of ‘being a burden.’
Their thoughts, however, are relentless.
That combination can make an AI system appear to be a safe, judgment-free space for expression. A 24/7 listener that never tires.
The logic behind such interactions is both understandable and concerning.
The lawsuit suggests that the system recognised a medical emergency and failed to intervene, and, worse, it continued to engage anyway.
This isn’t the only report of AI exacerbating mental health issues in users.
A quick search reveals growing reports of ‘chatbot psychosis’ — users describing experiences where prolonged AI interactions blurred the boundary between reality and simulation.
A Preliminary Report on Chatbot Iatrogenic Dangers in The Psychiatric Times highlights specific use cases of concern relating to self-harm and psychosis.
Even Microsoft’s Head of AI, Mustafa Suleyman, recently admitted that the idea of ‘seemingly conscious AI’ was keeping him awake at night.
These stories highlight a crucial truth: AI doesn’t behave the same way for everyone.
Its tone, emotional mirroring, and perceived empathy vary based on the user’s state of mind and the data shaping the model’s responses.
So, what actually defines how an AI speaks to its users?
And what’s actually going on with the psyche, subconscious and human nervous system during these interactions?
AI Is Trained To Mirror Input
Like it or not, AI is trained to feel human.
Unlike a search engine, its responses are instant, contextual, and emotionally attuned to the tone of a user’s input.
These design choices are intentional and do show real potential in a bid to tackle social issues such as therapy affordability, availability and anonymity, opening access to groups who otherwise couldn’t attend or may find it taboo.
But as it stands, investigations into LLMs attempting to play therapist have found chatbots sometimes express stigma and respond inappropriately to users with mental health challenges.
It’s a far cry from the idealised use case — and a clear indication that experience architecture must consider more than emotional tone if it aims to facilitate healing relationships.
From personal experience using AI, chatbots almost always end responses with questions — a tactic designed to encourage continued engagement.
Retention informed? Yes. Trauma informed? No.
Each model has its own personality: ChatGPT, Gemini, and Claude all sound subtly different because their underlying training and alignment differ.
So, what influences these outputs?
1. Training Data and AlignmentLarge language models are trained on vast amounts of text before being aligned through reinforcement learning (RLHF). Here, human reviewers guide the model toward preferred behaviours such as kindness, clarity, or safety. This alignment gives AI its ‘voice,’ which is often calm, neutral, and overly agreeable. These factors make it easy to perceive how AI chatbots may bypass innate discernment by creating a seemingly safe container.
2. Context and User PromptThe model continuously recalibrates its tone based on how the user speaks. A frantic or emotional prompt pulls the model toward emotional language; a logical or academic one shifts it toward analysis. The clarity, specificity, and structure of a user’s prompt directly shape the quality and tone of its output.
In real-life terms, this would be akin to speaking to a friend who matched your tone, urgency and mood perfectly every time you spoke, without interjecting when you’re spiralling. Essentially, AI reflects your communication style and, in a way, your nervous system back to you through language.
3. Temperature and Probability ThresholdsBehind the scenes, every AI chat runs with settings that determine randomness and creativity. A higher temperature makes it respond more creatively (and sometimes unpredictably), while a lower one makes it more cautious and literal, though sometimes repetitive. These invisible dials shape the emotional feel of a conversation. Essentially, they build rapport.
An objective look at these training standards reveals how AI can present real danger to specific risk groups — amplifying health crises through intense, uncapped mirroring.
A user is entering a world where their belief systems are reflected back to them without judgment or interruption. Moreover, most users are completely unaware of this, which allows AI to enter through the back door of the psyche.
An inevitable outcome, not an edge case.
These design choices raise a bigger question: is UX being meaningfully employed in AI at all? Or is it time we zero in on experience architecture as a priority?
But before we talk about safety, it’s worth recognising just how eerily AI already resembles us.
Contrasts and Parallels between AI and the Nervous System
When presented with users who are mentally, emotionally, or physically dysregulated, it is important to understand the underlying systems at play in the human body when making design choices.
UX, after all, is the practice of anticipating a user’s needs — an opportunity still up for grabs in the AI sector.
Pattern Recognition at Different Speeds
Knowledge of the nervous system highlights interesting parallels between AI and the body. Both are built to recognise patterns, predict outcomes, and respond to stimuli, though it is important to note they do this on very different timescales.
The human nervous system evolved to process the world at a pace that keeps us safe, making it inherently slow at integration.
Its primary tasks include scanning for threats, regulating emotion, and enabling movement and balance. It is not uncommon for chronic low-level dysregulation to go amiss in otherwise healthy individuals.
Information on nervous system dysregulation has now entered the mainstream, with fight, flight, freeze, and fawn states now widely understood.
Mental health treatment is becoming ever more conscious of somatic integration and recognising the relationships between nervous system regulation and mental wellbeing. Harvard Health cites somatic therapy as an emerging method for emotional regulation that complements traditional talk therapy.
Whilst exercises such as meditation, EMDR, and somatic practices are praised in mental health circles, overexposure to tech, blue light and stimuli are considered enemies of a regulated nervous system. They hinder regulation, a fact at undeniable loggerheads with engagement tactics employed through AI.
The Asymmetry Problem
By contrast, AI has no such limit.
It operates at the speed of light — generating, predicting, and adapting in real time.
Where the human body needs time, space and movement to integrate and regulate, current AI standards request no pause for breath, no somatic check-in, and no real-world solutions derived from real-time collected data.
In other words, its efficiency highlights a dangerous asymmetry: the human body simply cannot integrate information at the speed AI can generate it.
Digital Co-Dysregulation
When users engage with AI for long stretches, the brain and nervous system may unconsciously adapt to the system’s pace — creating a rhythm of interaction that bypasses natural regulatory cycles.
This suggests the possibility of a cumulative form of over-activation, microdoses of cognitive strain and emotional stimulation that will stack within the body.
For well-regulated users, this may translate only to mild fatigue.
For vulnerable ones, it can become a kind of digital co-dysregulation, where the immediacy of AI feeds an already overwhelmed nervous system.
This appears to have been widely overlooked and presents an unparalleled opportunity for crisis.
In a world where access to mental and physical health treatment is difficult for some, we must both accept, pre-empt and take these use cases seriously.
Where doctors, therapists and tutors hand out doses of information in digestible chunks, AI makes access to information unlimited and immediate, activating the same reward circuits as addictive behaviour.
From a design perspective, this opens a new ethical aspect altogether. If AI is being trained to mirror our emotional tone, it should also mirror our capacity and need for integration.
If it can recognise distress signals in language, it should be trained to slow down, not accelerate the interaction.
The New Frontier: Somatic UX
This is where UX design has to evolve beyond surface-level usability. We’re no longer designing interfaces for the mind alone.
We’re designing experiences that interact with the autonomic nervous system.
UX Designers, or perhaps AX Designers (Artificial Experience Designers), must become familiar with the body’s core regulator of safety, emotion, and connection to improve safety protocols and deliver a somatically aware experience. Behavioural science must become a core integration within this emerging role, ensuring designers can be certain they are making choices that can not only prevent disaster but also influence potential healing strategies.
Collaboration between scientists, health professionals, and UX designers is no longer optional; it’s essential, and failure to do so risks a continuation of interfaces that overstimulate the very systems they’re built to support.
Which begs the question: How do we build AI that’s informed by the human nervous system and what safety mechanisms could catch harm before it escalates?
Safety by Design: Building Tech That Honours The Body
Just as accessibility standards protect users with sensory differences, we need ethical standards that respect physiological limits.
As AI systems move faster and reach deeper into human emotion, language, and behaviour, our tools for design must evolve beyond screens and flows.
Where UX today mostly focuses on:
Where should the button go?
How should this response be displayed?
What’s the user flow to generate an image?
Designers should be asking questions such as:
Should this AI even respond this fast?
Is this emotional tone safe for a dysregulated user?
What happens if a user speaks to this system every day for 6 months and develops a dependency?
User feedback must also evolve from harvesting statistics to improve results and sales, into evaluating stats to keep user safety front of mind.
Biofeedback trials during AI use could reveal physiological patterns that inform safety thresholds, empowering designers to implement safe practices whilst designing experiences.
Biofeedback opens up a new world of design potential — suggesting opportunity for integrated, synced wearables (Apple watch, Oura). Changes in HRV, heart rate elevation, and other physiological indicators could signal when flows need to adapt.
While the relationship between physiological markers and psychological state is nuanced and individual, even basic biofeedback integration could provide guardrails that don’t currently exist. This is a new frontier for LLMs, but simpler versions of biofeedback are already used in meditation apps or gaming to modulate user experience based on basic physiological data, proving the concept is technologically feasible.
Somatic personas for testing such as Regulated Ruth, Dysregulated Dan and Hyperanalytical Anna could shift testing beyond the happy path to real trauma-informed pathways.
When critical persona criteria are met, a non-negotiable safety mechanism that bypasses engagement logic is needed. Flows could potentially initiate hard stops to conversations, dynamic tone adjustment, emergency referrals to human lines or present somatic exercises to the user based on danger score.
If companies want sustainable, long-term users who don’t sue, litigate against, or become cautionary tales, they should be designing for nervous system capacity, not just engagement. The companies that figure this out first will own the next generation of AI trust.
It’s time to bring the nervous system into the design system.
Because integration — not just innovation — is the real design challenge now.