Anthropic Splits Claude 5 Into Two Tiers: Mythos for Partners, Fable for Everyone Else

Anthropic is releasing two separate versions of its next-generation Claude 5 model at the same time. Claude Mythos 5 goes to a select group of trusted organizations, while Claude Fable 5 becomes available to the general public. According to Anthropic, Fable 5 is built so it cannot be used to carry out cyberattacks, a restriction that does not appear to apply to the Mythos tier reserved for vetted partners.

What happened

Anthropic is releasing Claude 5 in two separate versions at the same time. The first, Claude Mythos 5, goes to organizations Anthropic describes as trusted cyber partners. The second, Claude Fable 5, is the public release.

The key difference, according to Anthropic, is what Fable 5 cannot do. The company says the public model is built in a way that prevents it from being used to assist with cyberattacks. Mythos 5, by contrast, is being handed to a narrower group of vetted organizations, suggesting it carries fewer of those restrictions.

Why it matters

This is a notable structural decision. Most AI labs release one model and apply usage policies on top of it. Anthropic is baking the access tiers into the product itself, shipping two different models with different capability profiles.

For businesses, this creates a two-speed AI world at the model level, not just the pricing level. If your organization qualifies as a trusted cyber partner, you get a more capable tool. If you are a regular developer or business, you get a version with guardrails built in from the start.

There are real questions worth sitting with here:

• What exactly qualifies an organization as a trusted cyber partner, and who decides?
• How robust are the cyberattack restrictions in Fable 5 in practice?
• Will the capability gap between Mythos and Fable affect legitimate security research or red-teaming work?

The move also signals that Anthropic sees cyber capability as the sharpest edge of the risk profile for its most advanced models. That is worth paying attention to.

Our take

The two-tier structure is honest in one sense: Anthropic is admitting out loud that the same model cannot safely go to everyone. Most labs obscure that tension with vague acceptable-use policies and hope for the best. Publishing two distinct products at least puts the constraint on the label.

That said, the framing raises more questions than it answers right now. “Trusted organizations” is doing a lot of work in that sentence. Without a clear published standard for how a company earns Mythos access, this could easily become a system where large incumbents get the powerful tool and smaller teams get the neutered one, regardless of their actual safety practices.

For most of our clients, Fable 5 will be the only option on the table. The practical question is whether its restrictions create meaningful gaps for legitimate tasks like penetration testing, security audits, or vulnerability research. If Fable 5 refuses to engage with those workflows, teams will route around it with other models, which solves nothing.

Watch for third-party evaluations comparing what Fable 5 will and will not do in security contexts. That will tell you more than Anthropic’s framing will.

What to do about it

If your business uses AI for any security-adjacent work, test Fable 5 against your actual workflows as soon as access is available. Document where it refuses or degrades. That gives you a concrete picture of whether the restrictions affect your team, and whether you need to make the case for a different access tier or look at alternative models.

Source: WIRED · AI