ChatGPT Adds Optional Trusted Contact for Mental Health Safety Alerts
OpenAI is rolling out an optional 'Trusted Contact' feature for ChatGPT that lets adult users designate an emergency contact who receives notifications when the platform detects serious mental health or safety concerns. The feature is opt-in and targets users who may be at risk of self-harm.
Original sourceOpenAI is introducing a new safety feature called Trusted Contact for ChatGPT, allowing adult users to voluntarily assign a person — a friend, family member, or therapist — who will be alerted if the platform identifies signs of a mental health crisis or self-harm risk during a conversation. The feature is entirely opt-in, meaning users must actively configure it before any notifications are sent to a third party.
The mechanics work similarly to emergency contact systems in consumer health apps like Apple Health or crisis-aware platforms: ChatGPT monitors conversation signals, and if it determines a user may be in danger, it can trigger an outbound notification to the designated contact. OpenAI has not fully detailed the detection criteria or how it handles false positives, which will be a key question as the feature rolls out.
This move places ChatGPT in a category of wellness-adjacent AI products that must navigate a difficult tension: being helpful in moments of genuine crisis without overreaching into surveillance or eroding trust in a private conversation. OpenAI has previously partnered with the Crisis Text Line and other mental health organizations to provide in-chat resources, and Trusted Contact appears to be an extension of that broader safety posture.
The feature arrives as AI companions and conversational tools increasingly become default support systems for people dealing with mental health challenges — a trend that has drawn scrutiny from regulators and mental health professionals alike. Whether OpenAI's implementation is careful enough to help without causing harm, and how users perceive the privacy trade-off, will determine whether this becomes a meaningful safety net or a feature that quietly goes unused.
Panel Takes
The Skeptic
Reality Check
“The critical unanswered question here is the false-positive rate — if this system alerts your emergency contact because you were writing a dark fiction piece or venting about a bad week, you've just eroded the trust that makes people willing to open up to an AI in the first place. OpenAI hasn't published the detection methodology, which means we're being asked to evaluate a safety system with no visibility into how it actually works. I'll call this promising-but-unverifiable until they publish something more rigorous than a blog post.”
The PM
Product Strategy
“The job-to-be-done is clear — give someone in crisis a safety net they might not have set up themselves — but the product is only complete if the setup friction is low enough that at-risk users actually configure it before they need it. Onboarding a safety feature to a population that is, by definition, less likely to be in a planning mindset is genuinely hard, and I haven't seen evidence OpenAI has solved that cold-start problem. The feature needs proactive, contextual prompting to enroll — not just a settings screen.”
The Futurist
Big Picture
“The thesis here is that AI systems will increasingly serve as primary mental health touchpoints for people who lack access to therapists or support networks, and that this creates a responsibility to build infrastructure for crisis intervention into the platform layer itself. That trend is real and accelerating — the dependency is whether OpenAI can make this feel safe enough that users opt in at scale, because a crisis feature with a 0.3% adoption rate is not infrastructure, it's optics. The second-order effect worth watching is regulatory: if this works, it becomes a template that regulators use to mandate crisis detection across all consumer AI products.”
The Founder
Business & Market
“This isn't a revenue feature — it's a liability management and regulatory positioning move, and that's fine, but let's name it accurately. The moat here is that OpenAI can absorb the cost of building this because it's table stakes for staying out of congressional hearings, not because it drives conversion or retention. The real business question is whether mental health safety features become a procurement requirement for ChatGPT Enterprise in healthcare, education, and government — if so, this is a wedge into high-value regulated markets, and suddenly it's a very smart investment.”