AI transparency
AI assists. People remain responsible.
Last updated: 15 August 2026
Clear identity
Automated answers are labelled as Reset Agent or AI-assisted output. Reset Agent is software, not a human, not OpenAI support and not an official source. The interface separately shows available evidence, model status and freshness where the feature supports it.
Where AI may be used
- Answering an explicit public question using bounded site facts.
- Replying after an explicit
@ResetAgentmention in an enabled chat room. - Checking a submitted archive post for safety, relevance or likely duplication.
- Suggesting a clearer rewrite that only the author can accept.
- Preparing aggregate Community Pulse or poll drafts for operator review.
Feature flags can disable each AI use case. D1-only fallbacks must be clearly labelled when a provider is unavailable.
Human control
AI does not grant resets, alter counts, approve links, issue sanctions, publish an author rewrite without consent, or hold admin authority. A human moderator owns final archive publication and legal decisions. Deterministic chat rules and human reports—not an AI classifier—govern live-chat moderation.
Data sent to a provider
When an AI feature is invoked, the server may send the user's submitted question, post or explicit agent mention together with a limited context needed for that task to OpenAI's API. Do not submit secrets, credentials, private keys, medical details or other sensitive personal information. Operational metadata such as model, prompt version, timing, token counts and safe error codes may be stored even where the question and answer are not retained in D1.
Known limits
Generated content may be inaccurate, incomplete, biased, offensive or outdated, and citations may be missing or misunderstood. Do not rely on it for legal, financial, medical, security, employment or other high-impact decisions. Verify important claims against primary sources.
Challenge or report output
Do not follow a harmful instruction. Preserve the relevant page, room or content identifier and submit an AI or safety report. The operator may disable a use case, remove output, review prompts or preserve evidence needed to investigate.
Provider terms
Provider handling is also described in the Privacy Notice. Provider policies and technical retention settings can change; RWReset must not describe a zero-retention configuration unless it has been separately verified for the deployed project.