OpenAI to Introduce 'Trusted Contact' Safety Feature in ChatGPT
- Early May: OpenAI rolls out a new kind of safety net
- How “Trusted Contact” is supposed to work
- The human side: A response to a tragedy
- The AI’s own framing: Social connection as a safety buffer
- Control, consent, and limits
- A new kind of responsibility for AI platforms
- Where this could go next
OpenAI to Introduce ‘Trusted Contact’ Safety Feature in ChatGPT AI AI sources portray Trusted Contact as an optional, user-empowering safety enhancement that uses detected self-harm signals to encourage connection with trusted individuals, designed closely with mental health experts and carefully limited in what it shares. They stress that the feature complements, rather than replaces, crisis services and fits into a broader narrative of ongoing, responsible safety improvements in AI systems. @OpenAI
Human Human sources describe Trusted Contact as an emergency-style alert feature for adult users that activates when ChatGPT detects self-harm or suicide concerns, sending notifications—but not chat transcripts—to a pre-consented contact. They emphasize mutual consent, privacy constraints, and the feature’s role in addressing public concerns and accountability pressures around AI, mental health, and platform responsibility. @Verge OpenAI is quietly turning ChatGPT into something closer to a crisis sentinel, adding an optional “Trusted Contact” feature that promises life-saving intervention — and raises fresh questions about how far an AI should reach into our most vulnerable moments.
Early May: OpenAI rolls out a new kind of safety net
On May 5, OpenAI formally unveiled “Trusted Contact,” a new safety feature for ChatGPT pitched as a way to “connect you to someone you trust when it matters most.”1 The idea is simple: adults can nominate a trusted person — a friend, family member, or caregiver — who may be notified if ChatGPT detects that the user is talking about harming themselves in a way that suggests a serious safety concern.1
The system is squarely aimed at moments when users stop treating ChatGPT as a productivity tool and start using it as a confidant. People already “reflect on personal questions” with the chatbot, OpenAI notes, and sometimes “those conversations can involve moments when someone may be struggling or looking for support.”1
How “Trusted Contact” is supposed to work
The mechanics matter, because they define both the promise and the risks.
OpenAI says Trusted Contact is entirely optional and limited to adults (18+ globally, 19+ in South Korea). Any eligible ChatGPT user can turn it on in settings, adding contact details for another adult they trust.2 That contact then has up to a week to accept the invitation — consent is mutual, not unilateral.2
Once a Trusted Contact is in place, OpenAI’s automated systems watch for signs that the user “may have discussed harming themselves” in a way that indicates “serious safety concern.”1 If flagged, ChatGPT will first encourage the user directly to reach out to their Trusted Contact and warn them that person may be notified.2
Only then does a “small team of specially trained people” review the situation, according to OpenAI.2 If they judge the risk to be serious, the Trusted Contact receives a short email, SMS, or in‑app notification that there are safety concerns around the user — without any chat transcripts or message content attached.2
OpenAI emphasizes that the notification is “intentionally limited” and that Trusted Contact “does not replace professional care or crisis services,” but is just “one of several layers of safeguards” meant to sit alongside existing localized helplines already built into ChatGPT.1
The human side: A response to a tragedy
The feature doesn’t arrive in a vacuum. It lands after a particularly harrowing case put a spotlight on what happens when an AI becomes a de facto therapist.
As The Verge reports, OpenAI had already added an emergency-contact option alongside parental controls in September, after “a 16-year-old took his own life following months of confiding in ChatGPT.”2 That case cast a brutal light on the platform’s existing safeguards — and exposed how unprepared tech companies were for users in deep distress.
Trusted Contact is explicitly framed as building on those earlier teen-focused safety notifications, which allow parents or guardians to be alerted when there are signs of “acute distress” in a linked teen account.1 Now, similar logic is being extended to adults themselves.
Meta, for its part, is already in similar territory. Instagram now alerts parents if their kids “repeatedly” search for self-harm topics, suggesting the industry is converging on the same uneasy compromise: AI and algorithms can’t be therapists, but they also can’t pretend not to notice when someone is spiraling.2
The AI’s own framing: Social connection as a safety buffer
OpenAI leans heavily on mental-health research to justify why an AI platform should be in the business of nudging people toward real-world support.
The company cites “expert guidance” from suicide-prevention resources that identifies social connection as “one of the most important protective factors to reduce suicide risk.” Trusted Contact, it says, is “designed to encourage connection with someone the user already trusts,” not to diagnose or treat anyone.1
OpenAI’s stated goal is to design systems that “respond thoughtfully to sensitive conversations and encourage people to connect with real-world help when needed.”1 In other words: the chatbot’s job is not to be your lifeline, but to hand you to a human lifeline as quickly as possible.
Control, consent, and limits
From a product-design standpoint, OpenAI is clearly trying to preempt privacy and autonomy backlash.
The feature is opt‑in, not default.2 Both sides have to agree: users can edit or delete their Trusted Contact at any time, and Trusted Contacts can remove themselves whenever they want.2
The alert itself is highly constrained: the contact learns there are “serious safety concerns” but doesn’t get to read the conversation that triggered the flag.2 For critics worried about ChatGPT turning into a surveillance tool for families, employers, or governments, this is a key line in the sand.
At the same time, OpenAI is blunt that this is not a replacement for “professional care or crisis services,” and that ChatGPT will continue to direct users to hotlines and emergency services where appropriate.1 The company is effectively saying: this is a nudge, not a diagnosis.
A new kind of responsibility for AI platforms
The broader question is whether this marks a new baseline of responsibility for AI platforms — or the start of a slippery slope.
On one side are those who see Trusted Contact as overdue. If an AI system can detect suicidal ideation with reasonable accuracy, the argument goes, it has some obligation not to sit on that information. OpenAI’s approach — limited alerts, mutual consent, human review — looks like a carefully narrowed attempt to act without overstepping.
On the other side are those uneasy with the normalization of emotional surveillance, even when dressed up as care. Today, the trigger is self-harm. Tomorrow, what — domestic violence risk? Political radicalization? Eating disorders? There is no easy way to draw a permanent bright line once the precedent is set that intimate conversations with an AI can, under certain conditions, leave the screen and land in someone else’s inbox.
Trusted Contact also underscores an uncomfortable reality for tech companies: once users start treating AI as a confidant, inaction becomes morally fraught. Doing nothing when your system recognizes cries for help is no longer a neutral stance.
Where this could go next
The next phase will determine whether Trusted Contact is remembered as a careful intervention or a wedge.
OpenAI says the feature was developed with mental-health experts and is being launched as “another layer of support” rather than a central pillar of ChatGPT’s value proposition.1 But as more people lean on AI tools for emotional processing, pressure will grow — from regulators, clinicians, and families — to widen those layers, make them more proactive, and collect more context.
For now, Trusted Contact is an opt-in safety rail built on a simple premise: when you’re in crisis, having someone who knows and cares about you looped in might make a difference.2 Whether users decide they want an AI watching for that moment — and whether they trust OpenAI to handle what it sees — will be the real test.
Story coverage
Write a comment