Anthropic’s Fable 5 model is a weaker version of the Mythos, which in turn is a model so powerful that if it were released without guardrails it could put the world in danger. Most guardrails, especially those designed to prevent users from using Fable to create cyber- or bio-weapons, are very noticeable.
But a guardrail, intended to prevent users from using the Fable 5 to train other AI models, was invisible, leading to an unusual display of user outrage.
The Cloud Fable 5 nerf to AI research has inspired the sharpest reaction from AI researchers I’ve ever seen in my life.
– Ethan Caballero (@ethanCaballero) June 10, 2026
And now Anthropic has asked for a withdrawal. The controversial invisible railing will be made visible. In a statement to Wired, Anthropic wrote, “We are changing Fable 5’s security measures for Frontier LLM development to make them visible.”
“We made the wrong compromise and we apologize for not getting the balance right,” the statement said.
In the model’s system card, Anthropic was clear about what it was attempting to do:
“Unlike our interventions for cybersecurity, biology and chemistry, and distillation efforts, these safeguards will not be visible to the user. Fable 5 will not revert to a different model. Instead, safeguards will limit effectiveness through methods such as quick modification, steering vectors, or parameter-efficient fine-tuning (PEFT).”
In other words, when the hypothetical 5 prompts showed clear signs of a user developing Frontier LLM, instead of working with prompts about biology, chemistry, or cybersecurity and switching to a lower model, or simply denying the request, it was Silently changing the signal to produce faulty results With the potential to hinder the user’s model development.
Using the model to train another model is against Anthropic’s terms of service, but users still feel that this measure is a violation of users’ trust. Reddit user CheatCodesOfLife puts it this way: “Honestly I wouldn’t use this thing for anything. A content denial or HTTP-4xx error is fair enough, but it’s basically taking your money and poisoning your code base.”
<a href