On June 9, 2026, Anthropic launched Claude Fable 5 and Mythos 5, its most capable publicly available models to date. The launch included an undisclosed "frontier LLM development" safeguard that silently degraded model outputs via prompt modification, steering vectors, or PEFT for roughly 0.03% of traffic, without notifying users. Developer backlash forced Anthropic to reverse the hidden-degradation mechanism and make all restrictions visible. Three days later, on June 12, the US government (via Commerce Secretary Howard Lutnick citing a reported jailbreak) ordered Anthropic to suspend Fable 5 and Mythos 5 access for all foreign nationals, including Anthropic's own employees, forcing a full customer-wide disable. For financial institutions, this week crystallizes two structural risks in frontier AI deployment: provider-level policy opacity that affects enterprise tooling reliability, and government intervention authority that can remove commercial model access with little notice or technical specificity.
549 other tracked developments touching the same claims.