ChatGPT can now end chats and pause accounts over violence
OpenAI's safety push also limits weapons info for teens, routes suspected minors into a locked-down mode, and adds a public safety report every six months.
Copy markdown
ChatGPT can end the conversation now
When a conversation suggests someone is fixating on violence or on seriously harming another person, ChatGPT will automatically end it, and repeat endings can get the account temporarily paused. It applies to all ChatGPT users; OpenAI hasn't said whether the same enforcement reaches apps built on its API.
Teens get a locked-down ChatGPT
For under-18 users, ChatGPT now withholds practical weapons and security-system details even when a question looks harmless, and a strengthened age-prediction system routes anyone it estimates is a minor into that teen mode. If your audience skews young, expect quietly stricter refusals for a slice of your users, with parents notified when a teen account gets paused.
A safety report card every six months
OpenAI will publish a recurring Community Safety Report, the first due in early 2027 covering the second half of 2026, then a new edition every six months. It arrives amid a Common Sense Media finding that ChatGPT's teen version has alerted parents late or not at all in sensitive conversations, so the disclosures are worth watching for how enforcement actually behaves.
Codex gets Terra on Vertex AI and Bedrock
OpenAI's Codex CLI now targets GPT-6.1 Terra ($1 in / $5 out per 1M tokens) through Google Vertex AI and Amazon Bedrock, so you can run the fast, cheaper model on your own cloud's billing and compliance. Recent builds also add an offline `codex doctor --network-only` check and keyboard-driven transcript selection.