Sunday, 1 November 2026

20:10

Anthropic is expanding its text watermark to four more Claude models

Anthropic says it will watermark Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6 and Claude Sonnet 4.6, and points to the EU AI Act's December deadline as the reason. On Claude Opus 4.6 and Claude Sonnet 4.6, the watermark kicks in only at default settings, and a minor tweak to any one of temperature, top_p or top_k is enough to dodge it, according to The Decoder.

NoteFor anyone relying on the mark for compliance, Anthropic advises keeping the sampling settings at their defaults.

Source: The DecoderShare

20:10

Models & Labs

OpenAI will train models on Daybreak Blue session data from November 29

Starting November 29, OpenAI will use Daybreak Blue session data to train its models, according to an email to Trusted Access program members. Previously, this data was excluded from training even when users had the "improve the model for everyone" setting turned off.

NoteThe change has researchers worried about a repeat of the Buckmaster-Alpöge case, in which mathematicians uploaded drafts to OpenAI's Codex. OpenAI denied using their inputs for training, and after backlash it promised a separate opt-out.

Source: The DecoderShare