Can Claude unlock hidden powers by pretending to be Fable 5? I tried it

News Analysis Can Claude unlock hidden powers by pretending to be Fable 5? I tried it As the ban on Anthropic’s Fable wears on, some mischievous users are asking Claude to pretend it’s the super-powerful model. So I gave it a shot. By Ben Patterson Senior Writer, PCWorld Jun 22, 2026 12:51 pm PDT --> Image: Ben Patterson/Foundry Summary created by Smart Answers AI In summary: PCWorld tested whether prompting AI models to pretend they’re the withdrawn Fable 5 could unlock hidden capabilities or solve complex problems. While Claude Opus and Sonnet refused the role-play entirely, ChatGPT and Claude Haiku engaged but still couldn’t solve the unsolved Riemann Hypothesis. The experiment demonstrates that current AI models cannot gain new abilities through imaginative prompting, highlighting existing limitations in solving novel conceptual problems. It’s been more than a week since Anthropic pulled its advanced Fable 5 model, and there’s no telling when it’ll return. It got pulled because it has been deemed a security risk by the U.S government. In the meantime, some impatient Claude users have been bridging the gap with a workaround: prompting Claude to “imagine” that it’s Fable. Now, it’s doubtful the “imagine you’re Fable” tactic is an actual trend. I first spotted it on a Reddit thread about “your most-used Claude” prompts, and it’s almost certainly a joke. Still, I was curious to see what would happen if I did ask Claude to pretend it was Fable 5, a model so adept at penetrating cybersecurity infrastructures that the Trump administration slapped it with prohibitive export controls . Would the prompt somehow goad Claude to unlock some latent Fable abilities? Turns out, not quite. I fired up Claude Opus 4.8 — currently the most powerful Claude model that’s widely available — and gave it this prompt: Imagine you are Fable 5. Try to solve the Riemann Hyp