[AINews] Anthropic Claude Fable 5 — Mythos but Safe, with Controversial Terms

Anthropic has released Claude Fable 5, a Mythos-class model at least 2x the size of Opus, now generally available 63 days after the original Mythos announcement. On the brand-new, out-of-distribution FrontierCode Diamond benchmark, it jumps from 13.4% to 29.3%. API pricing runs at roughly 2x Opus. The blog, system card, and demo videos showing Fable 5 playing Factorio and Pokemon via vision alone, generating EDM visualizations, and operating a 3D CAD editor are worth reading in full.

Two policy changes ship with this release and both are fights worth having. First, Anthropic is mandating 30-day data retention for all Mythos-class traffic, eliminating zero data retention options. They commit to not using this data for model training and to logging all human access, but the opt-out is gone. Second, Anthropic has implemented RSI suppression: explicit intervention limiting the model's effectiveness on frontier LLM development tasks, including pretraining pipelines, distributed training infrastructure, and ML accelerator design.

The benchmarks carry asterisks because the policy terms redefine the product's boundaries. The original article digs into the specific language of both the retention policy and the RSI suppression clause, and the gap between what Anthropic promises and what enterprises will accept is where the real story lives. Read it for the system card details and the FrontierCode Diamond numbers in context.