Why AI companion safety needs clear boundaries and human support

An AI companion can offer a place to put a difficult day into words. Before using one, it is worth understanding what happens when a conversation becomes more serious than everyday reflection.

We are sharing a recent paper because that question deserves attention. This is MAX AI Companion's original editorial commentary, prepared with AI assistance. It is not a republication of the paper or a clinical review.

A paper worth reading

Joanna M. Streck and colleagues published a Comment in npj Digital Medicine on 22 September 2026 about preparing chatbots for distress and suicide-related messages in high-risk healthcare settings. Their example is Suzy, a chatbot used with adults receiving treatment for opioid use disorder.

The authors describe automated routing for different kinds of messages alongside a separate human review and clinical follow-up process. They also discuss expert testing as models change. Their recommendations come from a pilot implementation and require further validation.

The paper does not evaluate MAX. Its clinical setting and staffing arrangements should not be assumed to apply to a consumer companion app.

Read the original paper on Nature's website.

The questions we think every companion should answer

For us, the most useful response to this discussion is to make a product's limits easy to inspect. A person should not need to infer those limits from a reassuring tone.

Three questions are worth asking before relying on any AI companion:

• What kind of help does it offer? A conversation for reflection, a clinical service, and an emergency service have different responsibilities. The description should say which one you are using.
• Who, if anyone, can act when something is urgent? Showing a resource is different from connecting you to a person. If an app cannot monitor, contact anyone, or intervene, it should say so clearly.
• What happens to what you share? Check the privacy notice for storage, online processing, and any human access. Avoid assuming that a personal conversation stays entirely on your device.

These are questions about informed use. They are not a checklist that certifies an app as safe, and a polished answer is not evidence that a promised capability works.

What this means for MAX

MAX is a text-first companion for adults who want to reflect on everyday hard moments. It is not therapy, medical care, crisis support, or an emergency service.

MAX may surface crisis resources, but it cannot monitor you, contact emergency services or another person on your behalf, or intervene in real time. We are not claiming that MAX provides the paper's clinical monitoring or follow-up arrangements.

Conversation history and saved memory are local-first. Online AI features may involve server routing and third-party processing. Our Privacy Policy explains those details; our Safety and Boundaries page explains what MAX can and cannot provide.

If you or someone else may be in immediate danger, contact local emergency services or a crisis hotline right away. Do not wait for an AI conversation to provide help that needs real-world action.

We want readers to be able to make that distinction before they need it. The reason to share this paper is to encourage careful questions about AI companionship, not to suggest that a journal or its authors endorse MAX.

Source

Streck, J.M., Swendeman, D., Comulada, W.S. et al. Preparing AI chatbots to respond to patient distress and suicidality in high-risk healthcare settings. npj Digital Medicine 9, 715 (2026). https://doi.org/10.1038/s41746-026-03288-9.

The linked paper is open access under CC BY-NC-ND 4.0. This post discusses its ideas in original wording and does not reproduce its text or figures. The authors and publisher are not affiliated with this commentary and do not endorse MAX.