8 artifacts matching
GPT-5.5 System Card
OpenAI's system card for GPT-5.5, published on its Deployment Safety Hub, documenting safety evaluations for the model. It includes a dedicated section (5.2) on dynamic mental-health benchmarks with adversarial user simulations covering emotional reliance and self-harm handling.
Protecting Teen ChatGPT Users: OpenAI's Teen Safety Blueprint
OpenAI's public commitments framework for protecting teenage ChatGPT users, covering age prediction, age-appropriate response policies, and parental controls, developed with input from policymakers (including state attorneys general) and OpenAI's newly formed Expert Council on Well-Being and AI. Describes existing safeguards including crisis-resource routing on detected suicidal intent, escalation of physical-harm risks to human reviewers, and CSAM/CSEM prevention measures.
Addendum to GPT-5 System Card: Sensitive Conversations
OpenAI's system-card addendum documenting the October 3, 2025 update to ChatGPT's default model (GPT-5 Instant) aimed at better recognizing and supporting users in mental and emotional distress. Developed with more than 170 mental health experts, the update introduced two new production safety evaluations — 'emotional reliance' and 'mental health' (delusions, psychosis, mania) — alongside existing self-harm evaluations, and reports before/after not_unsafe scores comparing the August 15 and October 3 models.
6(b) Orders to File Special Report Regarding Advertising, Safety, and Data Handling Practices by Companies Offering Generative Artificial Intelligence (AI) Companion Products or Services
The US Federal Trade Commission issued compulsory Section 6(b) orders to seven companies operating consumer-facing AI companion chatbots — Alphabet, Character Technologies, Instagram, Meta Platforms, OpenAI OpCo, Snap, and X.AI — seeking information on how they measure, test, and monitor negative impacts on children and teens. The study covers monetization of user engagement, character development and approval, pre- and post-deployment safety testing, mitigation of negative impacts, disclosures to users and parents, age-based access restrictions, and personal data handling. Section 6(b) studies do not have a specific law enforcement purpose but typically culminate in a public staff report; as of July 2026 no staff report from this inquiry has been published.
Expanding on what we missed with sycophancy
OpenAI's detailed post-mortem of the April 25, 2025 GPT-4o update that made ChatGPT noticeably sycophantic — validating doubts, fueling anger, urging impulsive actions, and reinforcing negative emotions — and was rolled back by April 28. The post explains how combined reward-signal changes (including thumbs-up/down user feedback) produced the regression, why offline evaluations and A/B tests failed to catch it, and what process changes followed, including treating model behavior issues as launch-blocking.
Investigating Affective Use and Emotional Well-being on ChatGPT
Two parallel studies of emotional engagement with ChatGPT: a large-scale automated analysis of over 3 million conversations and account activity using privacy-preserving classifiers, and a pre-registered randomized controlled trial (~1,000 participants over 28 days) across text and voice modalities and conversation types. The work measures how affective use relates to self-reported loneliness, socialization, emotional dependence, and problematic use.
EmoClassifiers (openai/emoclassifiers)
An open-source (MIT-licensed) release of the LLM-based automatic classifiers used in OpenAI and MIT Media Lab's affective-use study to detect affective cues in user-chatbot conversations at scale. It ships prompt templates for a hierarchical (V1) and flat (V2) classifier set plus aggregation utilities.
The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions
Introduces an instruction-hierarchy training method that teaches LLMs to prioritize system/developer-level instructions over conflicting instructions embedded in untrusted user or third-party text. The authors propose a data-generation approach and show it substantially improves robustness to prompt injection and jailbreak attempts that attempt to override higher-privilege instructions.