Anthropic's Unreleased Mythos Model: Too Powerful to Ship Safely

Alongside the Opus 4.7 release, Anthropic made an unusual public admission: it has built a more capable model called Mythos that it has not released to the public due to safety concerns.

This is the first time a major AI lab has openly conceded that its best model cannot yet be deployed safely. Anthropic said Mythos passed internal benchmarks significantly above Opus 4.7 across coding, reasoning, and multimodal tasks — but that safety evaluation protocols were not yet satisfied.

What this means for the industry: The disclosure adds pressure on regulators and competing labs to define what "safe enough" means at the frontier. It also validates Anthropic's constitutional AI approach — building systems that can refuse release even when commercially valuable.

For enterprise AI buyers, Mythos represents what the near-term roadmap looks like once safety frameworks catch up to capability.