Anthropic's Fable model, a safety-focused AI, has faced significant hurdles since its June 9th release. Initial export controls by the US government restricted its use to US citizens, forcing Anthropic to temporarily pull the model. After negotiations, these controls were lifted, though Mythos remains restricted.

The author, a developer of the RNA-seq quantification tool salmon, found Fable unusable for a C++ to Rust rewrite project. The model flagged prompts related to biological data as safety concerns, even when the task was purely software engineering. Despite attempts to clarify the prompt's nature, Fable refused engagement and offered no detailed explanation for its rejection, unlike Opus 4.8, which successfully assisted with the rewrite.

This experience, shared by other users reporting rejections for simple queries, suggests Fable's safety classifier may be miscalibrated or rely on broad rejection lists. The author's second attempt with Fable, post-re-release and with stricter safeguards, also proved disappointing. The model's inability to handle technically focused prompts, even those within its purported strengths, renders it impractical for research tasks.