The U.S. government killed Anthropic’s most capable artificial intelligence (AI) models on June 12, 2026, and crypto markets betting on the company’s so-called private valuation immediately felt it.
Anthropic released Claude Fable 5 on June 9, 2026, billing it as the most capable publicly available model in the company’s history and part of a broader family called Mythos-class models. The release drew immediate complaints. Anthropic had embedded aggressive safety classifiers that monitored user prompts in real time and silently rerouted flagged queries to an older, weaker model, Claude Opus 4.8, without completing them with Fable 5’s full capabilities.
Three categories triggered the rerouting: cybersecurity tasks with offensive or exploitative potential, biology and chemistry queries touching dual-use research, and what Anthropic called “distillation,” meaning attempts to extract Fable 5’s capabilities to help train competing frontier AI systems. Anthropic disclosed that these triggers affected fewer than 5% of sessions on average, but users reported the classifier fired on benign inputs, including, in some cases, simple greetings.
Developers and researchers publicly accused Anthropic of quietly degrading the model without clear disclosure. By June 11, according to a Wired report, Anthropic had issued an apology, made the fallback system more transparent, and adjusted some of the restrictions around AI development tasks. Two days later, the situation escalated beyond anything the company could patch.
