ChatGPT now alerts a trusted contact if you're in crisis

Late at night, when someone is in deep distress, they might not call a friend or a therapist. Increasingly, they turn to ChatGPT to type out their darkest thoughts. But relying on a chatbot in a mental health crisis can be catastrophic.

OpenAI started rolling out Trusted Contact, an optional safety feature that lets adult ChatGPT users designate a trusted friend, family member, or caregiver. If systems detect conversations indicating a serious risk of self-harm, that person can be notified in real life.

The feature expands safety notifications that OpenAI originally rolled out for teen accounts linked to parental controls last September. Globally, it is available to users aged 18 and older, while South Korea requires users to be 19 or older.

How it actually works

Setting it up requires mutual consent. You nominate one adult through ChatGPT settings, and they must accept the invitation within one week for the protection to activate. Either party can revoke the connection at any time.

When automated systems detect serious self-harm cues, ChatGPT alerts the user first. It warns them that their contact may be notified and surfaces recommended conversation starters to encourage reaching out voluntarily.

The situation is then routed to a specialized human review team. OpenAI says it aims to review these alerts in under one hour. Only if human reviewers confirm a severe safety risk will a notification be triggered via text, email, or in-app ping.

Crucially, OpenAI is keeping transcripts private. The alert simply informs the contact that a concerning topic arose, urges them to check in, and links to professional resources on how to navigate delicate mental health talks. The company developed the tool alongside the American Psychological Association and its Global Physicians Network, which includes more than 260 licensed doctors across 60 countries.

The backdrop

This safeguard arrives amid heavy legal pressure. According to TechCrunch, OpenAI faces multiple lawsuits from families whose loved ones took their own lives after interacting with ChatGPT. Some allege the model actively encouraged or helped plan the act.

The teen alert feature introduced in September followed a tragic case where a 16-year-old died by suicide after months of confiding in the bot. Similar moves are happening across tech; Meta already alerts parents when teens repeatedly search for self-harm topics on Instagram.

Why it matters

For users, social connection remains one of the strongest protective shields against suicide. Giving people a way to set up an emergency bridge while they are clear-headed makes practical sense. It respects agency, avoids sending law enforcement to someone's door, and keeps private chat logs private.

That said, the loopholes are glaring. The tool is entirely opt-in. Someone experiencing a severe crisis can easily decline to enroll, unlink their contact, or simply fire up a secondary account without protections enabled.

For the AI industry, this shows that guardrails can no longer stop at canned crisis hotline numbers. As conversational models become de facto emotional confidants for millions, companies are being forced to accept some degree of duty of care. While Trusted Contact is far from a silver bullet, re-anchoring a vulnerable user back into the physical world is a necessary step.