AI Safety Playbook for Random Text Chats in 2025

1. Safe Mode evolves into a layered trust system

Safe Mode remains on for new and teen accounts, but the fall update introduces layered trust scores. Every participant now receives a dynamic rating that reflects behavior, verified identity, and community feedback. Higher scores unlock faster queues; lower scores automatically trigger stricter pre-screening and slower message limits.

Parents, educators, and moderators can access a Safety Dashboard inside Support — Safety & Privacy to customize Safe Mode rules for groups they manage. The dashboard includes recommended templates for classrooms, campus clubs, and creator-led communities so every cohort can pick the right safety posture in minutes.

2. Real-time classifier refresh

We retrained the language model that powers instant message checks. It now spots policy violations in 22 supported languages, even when text is obfuscated or uses coded slang. Confidence scores dictate whether the system issues a warning, a soft block, or an automatic escalation to human analysts.

Only suspected messages are relayed to moderation hubs, and they are deleted as soon as a decision is made. Learn more about our privacy approach in the Privacy Policy.

3. Human-in-the-loop escalation within 90 seconds

Automation spots problems fast, but human judgment is still key. Our moderation analysts now receive flagged sessions in a Kanban queue that classifies severity and context. Most cases reach an analyst within 90 seconds, and a final decision is delivered within four minutes. That is fast enough to protect the community while preserving accurate enforcement records.

Enterprise partners using Thundr Chat via Company — Partner solutions can plug these events into their trust dashboards, making audits painless.

4. Reporting experiences people actually use

We redesigned in-chat reporting to require just one tap. The new modal captures optional details with suggested prompts such as spam or harassment that take under six seconds to select. Reports feed back into the trust system, nudging engaged community members to the front of the queue and quietly rate-limiting accounts with unresolved incidents.

Users can download a summary of their submitted reports for 30 days under Support — Account & Billing, reinforcing transparency.

5. Coaching moments that prevent incidents

We believe safety starts before a violation ever occurs. The September release adds contextual nudges that pop up when someone rapidly skips matches, types hostile language, or receives repeated skips. These nudges offer community reminders and, when needed, a short cool-down period.

Expect to see more positive reinforcement too—when a conversation earns mutual thumbs up, both participants receive queue boosts and a prompt to save the match as a favorite room.