What Claude Fable 5 Is and Why Anthropic Split Its Flagship Model
Claude Fable 5 is Anthropic’s most powerful generally available AI model, built on the same Mythos-class base system but equipped with strict safety guardrails and automated fallbacks to block high‑risk uses across cybersecurity, biology, chemistry, and model distillation. Instead of a single frontier model, Anthropic has released two faces of the same engine: Claude Fable 5 for the public, and Claude Mythos 5 for a vetted circle of cyber defenders, critical infrastructure operators, and Glasswing partners. Technically, the two share capabilities; the difference is that Mythos 5 runs without the cyber restrictions that define Fable. This split shows Anthropic’s view that advanced exploit‑finding abilities cannot be handed to everyone by default. It also anchors the Claude Fable 5 release in a wider shift toward responsible AI deployment, where access control is treated as seriously as model training.
How Fable 5’s Safety Classifiers and Fallback System Work
Anthropic wrapped Claude Fable 5 in a stack of AI safety guardrails rather than stripping capabilities out of the underlying model. Dedicated classifiers sit in front of Fable 5 to detect offensive cyber tasks, sensitive biology and chemistry help, and attempts to distill its abilities into another model. When they trigger, Fable does not hard‑refuse; instead, the request is routed to Claude Opus 4.8, which has tighter restrictions on exploit development, mass data exfiltration, ransomware code, and other near‑certain abuse cases. Users are told when this handoff happens, turning the fallback into a visible part of the experience. According to The Hacker News, “fallback fires in under 5% of all sessions,” so for more than 95% of use, Fable behaves like an unrestricted Anthropic Mythos model while still blocking direct help with cyberattacks and other dangerous workflows.
Why Mythos 5 Stays Locked to Vetted Security Users
Anthropic’s decision to reserve Claude Mythos 5 for trusted users comes from what earlier Mythos variants already showed in testing: these systems can discover and exploit vulnerabilities at a level that could change the balance between attackers and defenders. Internal red‑team work on Mythos Preview found zero‑day weaknesses across major operating systems and browsers, including a 27‑year‑old bug in OpenBSD and a remote code execution exploit against FreeBSD’s NFS server. The company describes Mythos 5 as “the strongest cybersecurity model in the world,” and it is kept behind a trusted‑access program that includes cloud providers, chipmakers, security vendors, and financial institutions. Anthropic frames the risk clearly: handing such exploit‑writing ability to any user with an API key would offer serious uplift to attackers, so the unrestricted Anthropic Mythos model is treated as sensitive critical‑infrastructure technology rather than a standard developer tool.
A New Template for Responsible AI Deployment and Industry Norms
Instead of weakening its latest model for public release, Anthropic is experimenting with a new pattern: same core system, different safety envelope. Fable 5 shows that a high‑end AI can be widely accessible while still enforcing strict limits on offensive cyber operations and sensitive science. The Hacker News notes that in tests where Fable was set to block outright, the classifiers stopped progress on exploit development, reconnaissance, lateral movement, and defense evasion, even when evaluated against 30 public jailbreak techniques. At the same time, Anthropic accepts that universal jailbreaks cannot be fully ruled out; its goal is to make them slow and costly to find so they can be caught before large‑scale misuse. This tiered approach may become an industry template, treating access controls, monitoring, and fallback models as core parts of system design instead of afterthoughts bolted onto raw capability.
Pricing, Access, and How Tiering Will Shape AI Strategy
Anthropic priced Claude Fable 5 and Claude Mythos 5 at USD 10 (approx. RM46) per million input tokens and USD 50 (approx. RM230) per million output tokens, less than half the cost of the earlier Mythos Preview phase. For now, Fable 5 is available through the Claude API and included on Pro, Max, Team, and seat‑based Enterprise plans at no extra cost until June 22 before it moves to usage‑based billing. Mythos 5, while priced the same, remains on a trusted‑access track, which means many developers will adopt Fable 5 by default even though they share a base model. That parity may encourage enterprises to standardize on Fable 5 for most workloads and reserve Mythos 5 for designated security teams. Over time, this two‑tier structure could push organizations to design internal AI policies that distinguish between general productivity use and tightly governed cyber‑defense operations.
