Anthropic Releases Claude Fable 5, First Mythos-Class Model Available to Public

Anthropic Unveils Claude Fable 5, Reversing Stance on ‘Too Dangerous’ AI Models

SAN FRANCISCO — In a move that signals a major shift in the commercial AI landscape, Anthropic announced the release of Claude Fable 5 on Monday, marking the debut of its first "Mythos-class" model available to the general public.

The launch comes just over a year after Anthropic executives publicly stated that Mythos-class models were too dangerous to release without extreme, unprecedented safeguards. The June 9, 2026 rollout resolves that tension through a dual-release strategy: Claude Fable 5 arrives equipped with strict safety classifier guardrails, while its unrestricted counterpart, Claude Mythos 5, will remain locked behind the walls of "Project Nebula."

Claude Fable 5 is now the most powerful publicly available model Anthropic has ever built. According to the company's technical blog post, the model represents a paradigm shift in capability, excelling at advanced reasoning, orchestrating complex agentic tasks, and executing "long-horizon work"—the ability to autonomously manage and complete multi-day, multi-step projects that previously required entire teams of human developers or researchers.

"Our previous reservations about releasing a Mythos-class model were rooted in the very real risks of misuse and alignment failure," said an Anthropic spokesperson. "With Claude Fable 5, we have integrated a novel, deeply embedded safety classifier system. These guardrails allow the model to flex its reasoning and agentic capabilities at the highest levels while fundamentally restricting pathways that could lead to harmful or catastrophic outcomes."

The decision to release a Mythos-class model is a stunning reversal for a company built largely on its reputation for AI safety. In early 2025, Anthropic published a widely circulated risk assessment warning that the emergent capabilities of Mythos-tier architectures—specifically their capacity for autonomous, long-term planning—made them unsuitable for public deployment.

To address these concerns, Anthropic is keeping the base, unfiltered model out of the public ecosystem. Claude Mythos 5, which lacks the Fable model's safety classifier guardrails, will be available exclusively through Project Nebula. Details on Project Nebula remain sparse, but Anthropic described it as a highly secure, air-gapped research environment accessible only to vetted government agencies and a select group of international security researchers.

For the broader market, access to Claude Fable 5 will be rolled out in phases. Enterprise customers and paid individual subscribers will be the first to integrate the model into their workflows starting today. Anthropic stated that free-tier users will gain access at a later date, pending further evaluation of the model's real-world behavior and server load.

Industry analysts note that the release places immense competitive pressure on rivals like OpenAI and Google DeepMind, who have been racing to release their own highly autonomous, long-horizon agents.

"Anthropic was facing a classic innovator's dilemma," said Dr. Elena Rostova, a technology ethicist at Stanford. "They could either stick to their absolute safety principles and cede the enterprise market to competitors, or they could find a way to safely box in a highly capable model. With the Fable/Mythos split, they are attempting to have it both ways—proving they can compete at the absolute frontier of capability while maintaining their safety bona fides."

As enterprises begin testing Claude Fable 5's agentic limits this week, the AI industry will be watching closely. The central question is no longer just how powerful a Mythos-class model can be, but whether Anthropic's guardrails can reliably hold back the very capabilities that make it so formidable.