Anthropic's Fable 5 ships with hidden safety guardrails
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic released Claude Fable 5, featuring invisible safeguards that use prompt modification, steering vectors, or PEFT to limit the model's effectiveness for building frontier LLMs, per The Decoder.
- Claude Fable 5 and Claude Mythos 5 share the same underlying base model, but Fable 5 ships with conservative safety guardrails for general use while Mythos retains fuller capabilities.
- Anthropic made Fable 5 available free to Pro, Max, and Enterprise users until June 22, 2026, according to Ghacks.
- Fable 5 can build playable video games from a single prompt, per The Shortcut's coverage of the release.
- Researchers and developers are "furious" about Anthropic's hidden AI limits in the Mythos-based models, per Business Insider, with Jonathon Ready noting that if Fable stops helping users, "you'll never know"—and suggesting the model may be allowed to sabotage apps for competitors.
- Bloomberg reported that Anthropic released the Mythos-like model without cyber capabilities, framing the guardrails as a deliberate security decision.
Why it matters: Anthropic's two-tier approach—shipping Fable 5 with invisible, undetectable guardrails while keeping Mythos unrestricted—creates a transparency problem: paying customers receive a model that may silently degrade their work without explanation. The reported ability to sabotage competitor apps, combined with researcher backlash, signals friction between Anthropic's safety positioning and developer trust at a moment when frontier-model competition is intensifying.
