analiza aiqbench

OpenRouter, an aggregator of AI model inference, is adding in-flight screening of prompts and responses against known-CSAM hashes, plus classifier-based flagging of suspected new material that goes to human review. The notable tensions for AI readers are the trade-off between content-safety scanning and zero-data-retention promises, and how far classifier flags can lead to retention, since the policy states a flag is not a determination of CSAM. Whether the provider's scope and the legal reporting obligations hold up in practice will be the real test.