Anthropic has officially launched its next-generation AI, splitting it into two distinct models: Claude Fable 5 and Mythos 5.
This move establishes a formal two-tier access system. Claude Fable 5 is designed for the general public but comes with built-in 'guardrails' to prevent misuse, especially in cybersecurity and biological research. In contrast, the unrestricted Claude Mythos 5 is reserved for a select group of vetted organizations through a 'trusted access' program. The pricing reflects this premium capability, set at exactly double that of its predecessor, Opus 4.8, though it's a significant 60% reduction from the initial pilot program price.
So, why this cautious approach now? The decision was shaped by a rapid succession of events over the past two months. First, just days before the launch, the White House issued an executive order requiring government review of powerful frontier models before their public release. This policy move directly validates and encourages a controlled access model like Anthropic's.
Second, Anthropic's own research, shared right before the announcement, revealed that Mythos 5 could turn newly discovered software vulnerabilities into working exploits in mere hours. This starkly highlighted the potential for misuse, making a fully open release seem irresponsible. Reports of the NSA allegedly using Mythos for offensive cyber operations further amplified these national security concerns, reinforcing the need for strict controls.
Finally, the groundwork for this strategy was laid in April with 'Project Glasswing,' a pilot program that gave a limited number of partners access to an early version of Mythos. This trial not only demonstrated the model's immense power but also helped Anthropic understand the necessity of a safer, more broadly available version, which has now become Fable 5. This entire sequence—from internal research and government directives to pilot program feedback—created the perfect conditions for today's dual-model launch, balancing innovation with a profound sense of responsibility.
- Frontier Model: Refers to the most advanced, powerful AI models currently in existence, possessing capabilities that can present significant societal risks.
- Trusted Access: A system where access to a powerful technology is restricted to carefully vetted users or organizations that have been deemed trustworthy.
- Exploit: A piece of code or a sequence of commands that takes advantage of a bug or vulnerability in a software system to cause unintended or unanticipated behavior.