ScatterLab AI Ethics: Principles, Chatbot Ethics Checklist, and Abuse-Response Policy

Key Findings

• — Frames its ethics principles explicitly around fostering 'intimate relationships' between users and its AI companion product, rather than generic AI-safety principles
• — Describes an abuse-detection and classification model that screens conversation turns for sexual, aggressive, or biased content before a response is generated, developed after reviewing Iruda 1.0's prior failures and academic literature on abuse in AI chatbots
• — Reports semi-annual random-sampling audits of the chatbot's 'safe utterance rate', with results across periods ranging from 99.56% to 99.85% against a stated 99% target; states that if a check falls below target, the abuse-detection and dialogue models are retrained and re-tested within three months
• — Documents that Iruda 1.0 was shut down approximately three weeks after its December 2020 launch due to inadequate personal-data-consent handling and discriminatory speech, followed by a year-long rebuild (database reconstruction, retrained language model, added abuse-response systems) before Iruda 2.0's closed beta launch in January 2022

Methodology Notes

A company-maintained ethics/policy hub rather than a single dated publication; the top-level ethics-principles page states a last-updated date of August 26, 2022, while the abuse-response policy section documents dated periodic safe-utterance-rate check results. Self-reported audit methodology (random sampling of chatbot utterances, semi-annual cadence) with no independent verification.

Sources