1. Anthropic reportedly adds text watermarks to four more Claude models, citing the EU AI Act's December deadline. But Claude Opus 4.6 and Claude Sonnet 4.6 carry the mark only when sampling settings such as temperature stay at their defaults, so an API user who changes one setting reportedly gets unmarked text. The report includes no independent test of the gap. [Unverified] primary
2. On New York's law barring AI companions for minors that flatter, suggest they are human, or promote self-harm: the governor signed a version that delays the effective date to June 1, 2027. [Unverified] primary