OpenAI Updates ChatGPT for Context-Aware Safety in Sensitive Talks

OpenAI has introduced significant updates to ChatGPT aimed at improving its ability to handle sensitive conversations where risk may emerge gradually. Announced on May 14, 2026, these changes enable the AI to better identify subtle patterns of distress or harmful intent by analyzing context across multiple interactions, rather than isolating messages. This advancement is part of OpenAI’s ongoing efforts to enhance safety in scenarios involving self-harm, suicide, or violence.

One of the key features rolled out includes “safety summaries,” which are short, factual notes capturing safety-relevant context from prior conversations. These summaries are narrowly scoped, stored temporarily, and designed to improve the model’s responses in high-risk situations. For example, if a user shows signs of distress over multiple chats, the summaries help the AI connect the dots and escalate caution appropriately—whether by refusing certain requests, de-escalating the conversation, or directing users to safer alternatives.

According to OpenAI, this update builds on over two years of collaboration with psychiatrists, psychologists, and safety experts. Testing showed notable improvements: in single high-risk conversation scenarios, safe-response performance improved by 50% in suicide and self-harm cases, and by 16% in harm-to-others situations. Across multiple conversations, performance gains were even higher, with a 52% improvement in harm-to-others cases and 39% in self-harm scenarios when using GPT-5.5 Instant, the current default model in ChatGPT.

Why Context Matters

OpenAI emphasized that context is often critical in sensitive interactions. A seemingly benign request might take on a different tone when viewed alongside earlier signs of distress. For example, a user asking generic questions about medications might signal deeper concerns if prior messages point to suicidal ideation. The updated model is trained to recognize these connections and prioritize safety in its responses.

The focus of this work has been on acute scenarios involving self-harm or harm to others, where early intervention can be life-saving. OpenAI’s safety summaries are not intended for personalization or long-term memory, but rather as a targeted tool for rare, high-risk situations.

Building on Broader Safety Efforts

This update is part of a larger initiative by OpenAI to make ChatGPT safer and more responsible over time. Earlier updates in October 2025 and January 2026 introduced measures like age prediction to reduce exposure to sensitive content for minors, parental controls, and safety routing systems that direct risky prompts to models optimized for safer outputs. Additionally, the company launched the “Trusted Contact” feature on May 7, 2026, which allows adult users to nominate a person who can be alerted if ChatGPT detects serious safety concerns.

These layered interventions reflect OpenAI’s shift toward longitudinal risk detection, where harm signals are identified and addressed over time rather than in isolated exchanges. The company has also increased transparency by publishing detailed evaluations of its safety performance metrics. For example, safety summaries received average relevance and factuality scores of 4.93 and 4.34 out of 5, respectively, in internal reviews.

What’s Next

While the current updates focus on self-harm and harm-to-others scenarios, OpenAI is exploring whether similar safety mechanisms could apply to other high-risk areas, such as cybersecurity or bioethics. Any expansion will include rigorous safeguards and expert collaboration, the company said.

As AI systems like ChatGPT become more deeply integrated into daily life, the ability to detect and respond to evolving risks will remain a critical challenge. For now, OpenAI’s updates mark a meaningful step forward in making conversational AI both more aware and more responsible in sensitive situations.

Image source: Shutterstock