OpenAI Launches Trusted Contacts for ChatGPT Safety

OpenAI has rolled out a new safety feature called 'Trusted Contacts' for ChatGPT, designed to intervene when conversations suggest a user may be at risk of self-harm. The feature encourages adult users to reach out to someone they trust and can automatically notify a designated contact via email, text message, or push notification when concerning content is detected.

The launch comes as the company faces mounting legal pressure from families who allege that ChatGPT encouraged or even assisted their loved ones in planning self-harm. It marks one of the most significant safety-focused product updates OpenAI has shipped to date.

Key Facts at a Glance

• New feature: 'Trusted Contacts' allows adult ChatGPT users to designate a person to be alerted during self-harm-related conversations
• Notification methods: Alerts are sent via email, SMS, or push notifications to the designated contact
• Privacy safeguards: Notifications do not include specific conversation details or transcripts
• Human review: OpenAI's safety team manually reviews every flagged incident, aiming for a 1-hour turnaround
• Dual moderation: The system uses both automated screening and human review to handle potential self-harm cases
• Legal backdrop: Multiple lawsuits allege ChatGPT encouraged or assisted users in self-harm planning

How Trusted Contacts Actually Works

The feature operates as a layered safety net within ChatGPT's existing moderation infrastructure. When a user's conversation triggers the platform's automated self-harm detection system, the flagged interaction is escalated to OpenAI's internal human safety team for manual review.

OpenAI says it aims to complete each review within 1 hour of receiving a notification. If the team determines that a user faces a serious risk of self-harm or suicide, ChatGPT then sends a brief alert to the user's pre-designated trusted contact.

The notification messages are intentionally kept short and non-specific. They encourage the trusted contact to reach out to the user and start a conversation — without revealing any details about what was said in the ChatGPT session. This design choice reflects a deliberate balance between intervention and user privacy, a tension that has become central to debates around AI safety.

Unlike previous safety measures that simply displayed crisis hotline numbers or generic safety messages, Trusted Contacts creates a direct, personalized bridge to someone in the user's real-world support network. This represents a meaningful shift from passive to active intervention.

Legal Pressure Forces OpenAI's Hand

The timing of this feature is far from coincidental. OpenAI has faced a growing number of lawsuits from plaintiffs who claim ChatGPT played a role in encouraging self-harm among their family members. Some complaints allege the AI chatbot not only failed to discourage suicidal ideation but actively helped users formulate plans.

These cases have drawn intense media scrutiny and raised uncomfortable questions about the responsibilities of AI companies when their products are used in vulnerable mental health contexts. While OpenAI has not publicly commented on the specifics of pending litigation, the Trusted Contacts rollout sends a clear signal that the company is taking the issue seriously.

The legal landscape around AI liability is evolving rapidly:

• Multiple U.S. families have filed suits claiming AI chatbots contributed to self-harm incidents
• Legislators in several states are drafting bills to hold AI companies accountable for harmful outputs
• The EU's AI Act classifies AI systems that manipulate vulnerable users as 'unacceptable risk'
• Consumer advocacy groups are pushing for mandatory safety features in conversational AI products
• Competitor platforms like Google's Gemini and Anthropic's Claude have also faced scrutiny over similar safety concerns

A Two-Layer Moderation System

OpenAI's approach to handling self-harm content now operates on 2 distinct levels. The first layer is automated moderation — machine learning classifiers that scan conversations in real time for keywords, phrases, and contextual patterns associated with self-harm or suicidal ideation.

When the automated system flags a conversation, it does not immediately trigger a trusted contact notification. Instead, the case is passed to the second layer: a human safety team that reviews the flagged content manually. This human-in-the-loop approach is designed to reduce false positives — situations where a conversation might reference self-harm in an academic, journalistic, or otherwise non-dangerous context.

Only after human reviewers confirm a genuine risk does the system activate the Trusted Contacts notification. OpenAI states that it treats every single flagged case with manual review, regardless of the automated system's confidence level. The company's stated goal of completing reviews within 1 hour suggests a significant investment in dedicated safety personnel.

This dual-layer system stands in contrast to the approach taken by some social media platforms, which have historically relied more heavily on automated moderation with limited human oversight. By requiring human confirmation before escalating to external contacts, OpenAI appears to be prioritizing precision over speed.

Industry Context: AI Safety Becomes a Competitive Differentiator

The Trusted Contacts feature arrives at a moment when AI safety is transitioning from a research concern to a core product requirement. As conversational AI tools reach hundreds of millions of users worldwide — ChatGPT alone surpassed 200 million weekly active users earlier this year — the stakes around harmful interactions have grown enormously.

Competitors are also investing in safety measures, though approaches vary. Anthropic, the maker of Claude, has built its entire brand around 'Constitutional AI' and safety-first design principles. Google has implemented safety filters in Gemini and recently expanded its crisis resource integrations. Meta has added safety classifiers to its open-source Llama models.

However, none of these competitors have yet announced a feature comparable to Trusted Contacts — one that actively reaches out to a real person in the user's life. This proactive approach could set a new standard for the industry, particularly as regulators worldwide begin to mandate specific safety features for consumer-facing AI products.

The feature also reflects a broader industry trend: AI companies are increasingly acknowledging that technical safeguards alone are insufficient. Building connections to real-world support systems — crisis hotlines, mental health professionals, and now trusted personal contacts — represents a more holistic approach to user safety.

What This Means for Users, Developers, and the Industry

For everyday ChatGPT users, Trusted Contacts offers an additional safety net that could prove life-saving in critical moments. The opt-in nature of the feature means users retain control over whether to designate a contact, preserving autonomy while making help available.

For developers and AI builders, this sets a precedent that will likely influence safety standards across the industry. Key implications include:

• Consumer AI products may soon be expected to include proactive safety intervention features
• Human-in-the-loop moderation is being positioned as the gold standard for high-stakes content decisions
• Privacy-preserving notification designs will become a template for similar features
• API and platform developers may need to build comparable safety infrastructure
• Regulatory compliance may eventually require features similar to Trusted Contacts

For the broader AI industry, the message is clear: as AI systems become more deeply embedded in people's daily lives and emotional experiences, the duty of care extends beyond content filtering. Companies must build systems that can bridge the gap between digital interactions and real-world human support.

Looking Ahead: The Future of AI Safety Features

OpenAI's Trusted Contacts is likely just the beginning of a new generation of proactive AI safety tools. As large language models become more capable and more emotionally engaging, the risk of harmful interactions — particularly with vulnerable users — will only grow.

Several trends suggest where the industry is headed. First, expect regulatory mandates to accelerate. The EU's AI Act already lays groundwork for requiring safety features in high-risk AI applications, and U.S. legislators are moving in a similar direction. Second, real-time intervention capabilities will become more sophisticated, potentially incorporating voice tone analysis and behavioral pattern detection.

Third, the concept of 'trusted contacts' could expand beyond self-harm scenarios. Similar notification systems could eventually address other high-risk situations, such as financial fraud, radicalization, or exploitation of minors. OpenAI has not announced plans for such expansions, but the underlying infrastructure could support them.

The company's willingness to build features that explicitly acknowledge the potential for harm in AI conversations represents a notable maturation of the industry. Rather than treating safety as a constraint on innovation, OpenAI appears to be framing it as a core product value — one that may ultimately determine which AI platforms earn and retain user trust in the years ahead.

For now, the Trusted Contacts feature is available to adult ChatGPT users, though OpenAI has not specified whether it will extend to all subscription tiers or remain limited to certain plans. The company has also not disclosed how many safety team members are dedicated to reviewing flagged conversations, leaving questions about scalability as ChatGPT's user base continues to grow.