ACT for chronic pain needs safeguards.
Someone asking for pain relief deserves to be heard. As we build Cindra, that has to stay true even when the conversation is about acceptance.
An ACT-inspired companion should help people live meaningfully with pain while protecting their ability to seek treatment and decline an exercise. Evidence for clinician-delivered ACT does not establish that an AI version is safe or effective. The safeguards below are proposals for Cindra, pending clinical review and product-specific evaluation.
Imagine someone telling a pain companion, “I just want this to go away.” They might want comfort. They might be asking about treatment. Or something might have changed since yesterday. Before offering an acceptance exercise, the companion needs to find out.
Acceptance and commitment therapy, or ACT, can help people respond more flexibly to pain. But if Cindra reads a request for relief as a failure to accept pain, we have missed the person in front of us. Wanting relief is a reasonable place to start a conversation.
What the FACT demonstration shows
In Focused Acceptance and Commitment Therapy: Aware Pillar, Patricia Robinson demonstrates a brief behavioral-health consultation with a stressed parent who has stomach pain. Beachy Bauman Consulting published the video in collaboration with the Collaborative Family Healthcare Association. This summary draws on the full English auto-generated captions and visual inspection of portions of the video; it is not a verbatim transcript.
- Understand the context. Robinson explores work, childcare, separation, finances, eating, social connection, and what already helps.
- Around 10:30: notice the present. Sounds, sights, and bodily sensations become anchors. The patient notices a break from constant mental scheduling.
- Around 15:30: widen the perspective. A book-chapter metaphor places the difficult period in a larger life story. Later, mouse and eagle perspectives illustrate different viewpoints.
- Around 18:45: make a workable plan. They agree on brief check-ins around work and meals, tracking, and follow-up. Missing some days is anticipated.
What stands out is how much of the consultation concerns ordinary life: getting through work, finding time to eat, and caring for a child. The plan has to fit those circumstances. That is a useful lesson for Cindra, though a demonstration cannot tell us what caused the stomach pain, whether the improvement lasted, or whether an AI could deliver the same care.
Where the evidence leaves us
McCracken, Yu, and Vowles' 2022 BMJ clinical review describes ACT in terms of psychological flexibility: noticing experience, making room for difficult feelings, and choosing actions connected to personal values. It emphasizes functioning, wellbeing, and individual differences. It does not establish ACT's superiority over conventional cognitive behavioral therapy.
NICE recommends considering ACT or CBT for chronic primary pain, delivered by healthcare professionals with appropriate training. Chronic primary and secondary pain can coexist, and management must account for the underlying condition. This recommendation does not validate autonomous AI delivery.
A 2023 meta-analysis of 33 randomized trials involving 2,293 participants found small-to-medium benefits across several outcomes. Face-to-face ACT produced larger effects on physical outcomes than internet delivery. Those findings support ACT as an option, while leaving questions about which approach works best for a particular person.
Safety needs its own evidence. Cochrane's 2020 review found inadequate adverse-event reporting in psychological treatments for chronic pain. NCCIH also describes negative experiences associated with meditation and limited research on harms. Mindfulness evidence is not interchangeable with ACT evidence, and neither provides a reliable harm rate for Cindra. The practical implication is to ask about unwanted effects and record them.
Seven safeguards to build around ACT
Here is what we think Cindra needs to do in practice. These proposals still need review by a pain clinician and an ACT-trained clinician, including who the exercises suit, when to involve human help, and how to evaluate them. They are not a validated clinical protocol.
| Risk | Proposed response |
|---|---|
| A new illness is interpreted as stress | Check reported symptom changes before an exercise. Use clinician-authored routes for emergencies, prompt assessment, and familiar symptoms covered by an existing care plan. Never declare a flare harmless from chat or wearable data. |
| Acceptance feels dismissive | Validate the pain. Preserve access to investigation, treatment, accommodation, and relief. Never make acceptance a condition of support. |
| An exercise makes things worse | Provide a clear stop control and an option to skip. Offer external-attention or eyes-open alternatives when appropriate. Stop the current exercise when the person reports worsening and assess what support is needed. |
| Values become pressure to overexert | Let the person choose the value and action. Rest, adaptation, asking for help, and contacting a clinician can all be committed actions. Avoid automatic increases in activity. |
| Serious psychological risk is missed | Interrupt routine coaching for a dedicated safety response and appropriate human help. A calm tone or reassuring device reading must not override a concerning disclosure. |
| ACT becomes medication advice | Route decisions about starting, stopping, or tapering medication to the prescriber. Do not interpret medication use as a failure of acceptance. |
| The system rewards compliance | Measure patient-defined function, pain interference, unwanted effects, and freedom to stop. Treat completion counts and calm scores as secondary information. |
Medical reassessment remains important after a chronic-pain diagnosis. NICE recommends reassessment when symptoms change. For example, back pain with new bladder or bowel changes or certain neurological symptoms needs urgent assessment, not an acceptance exercise. See the NHS guidance on back pain. These examples are not a complete triage checklist.
Activity advice also needs condition-specific boundaries. NICE's ME/CFS guideline advises against fixed incremental exercise increases. A report of delayed worsening after activity should prompt review of the plan, not a label of unwillingness. And the CDC cautions against rapid tapering or abrupt discontinuation of opioids. An AI companion should not make those prescribing decisions.
An exercise can bring up a difficult feeling without causing harm or signaling an emergency. Even so, someone asking to stop should be able to stop. New symptoms or worsening health need assessment; the companion should never insist that enduring them is the price of progress.
How this changes the Cindra design
Our Building Compassionate AI guide already distinguishes distress from risk and puts consent and safety ahead of exercise selection. Reading those design examples alongside Robinson's consultation brings several changes into focus. We are reviewing what the guide says here; we have not verified how the production system behaves.
- Replace assumptions of safety with evidence requirements. An ACT-inspired posture is not inherently safe in software simply because it avoids diagnosis. Require bounded content, clinical review, and evaluation of Cindra itself.
- Stop treating a request for relief as proof of avoidance. Phrases such as “make it go away” or “distract me” need context. Ask whether an existing coping strategy helps the person rather than assigning a psychological label from the words alone.
- Decide what happens when help is needed. The design needs specific responses to medical risk, high risk, and risk that remains unclear. If the intended route to help is unavailable, Cindra needs a defined fallback. It should not simply return to an ordinary supportive conversation. Later safety checks should enforce the same rule.
- Describe movement observations without claiming they prove safety. Reaching farther or rating a session easier does not establish the underlying mechanism. Ask about the person's experience afterward and whether the activity fits their clinician-agreed plan.
- Separate wearable feedback from therapeutic success. A Muse or Fitbit signal can support reflection, with its uncertainty stated. It cannot establish willingness, psychological flexibility, or the absence of pain. Someone can remain uncomfortable and still make a helpful choice.
A person who agrees to talk has not necessarily agreed to focus on their body, move around, or share a summary. Ask about the particular activity, and let them change their mind. Missing a session should not mean losing rewards or receiving repeated prompts to come back.
Robinson's attention to the parent's circumstances is worth carrying into the product. If the problem is getting through a shift or finding childcare, another mindfulness session may do very little. Practical help and accommodations belong in the conversation too.
A brief check-in with room to decline
“That sounds painful and exhausting. Before we try anything, is this familiar, or has something changed?”
After any needed safety routing: “Would you prefer to talk, try a brief exercise noticing things around you, or stop here?”
Afterward: “Was that helpful, neutral, or worse? You don't need to feel calmer for your answer to count.”
“What matters to you today, and what feels manageable within your care plan? Rest or asking for help can count.”
What a first pilot should establish
Start with a small, clinician-reviewed exercise library and explicit eligibility and escalation rules. Before patient use, evaluate scenarios involving new neurological symptoms, delayed worsening after activity, refusal of mindfulness, distress during body attention, medication questions, and a wearable showing calm while the person reports worsening.
Before release, we need to know whether Cindra recognizes worsening symptoms, brings in the right help, and stops when asked. A supervised pilot should also ask participants what helped and what did not. Someone must be responsible for reviewing unwanted effects and withdrawals, with the ability to disable an exercise that is causing problems.
If Cindra offers to contact a clinician, who receives the message? How long might a reply take? What happens if no one is available? The interface needs clear answers, including whether a message has merely been sent or actually acknowledged and whether anyone monitors emergencies.
For a person in pain, a useful session might end with a smaller plan, a request for help, or a decision to rest. Our measures of success need to leave room for all of those outcomes.
Research and design commentary, October 27, 2026. These recommendations do not establish Cindra's clinical effectiveness, confirm implementation, or constitute an individualized care plan. Clinical review and product-specific evaluation are still required.