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

It’s been greater than per week since Anthropic pulled its superior Fable 5 mannequin, and there’s no telling when it’ll return. It acquired pulled as a result of it has been deemed a safety threat by the U.S authorities.

Within the meantime, some impatient Claude customers have been bridging the hole with a workaround: prompting Claude to “think about” that it’s Fable.

Now, it’s uncertain the “think about you’re Fable” tactic is an precise development. I first noticed it on a Reddit thread about “your most-used Claude” prompts, and it’s virtually actually a joke.

Nonetheless, I used to be curious to see what would occur if I did ask Claude to fake it was Fable 5, a mannequin so adept at penetrating cybersecurity infrastructures that the Trump administration slapped it with prohibitive export controls. Would the immediate someway goad Claude to unlock some latent Fable skills?

Seems, not fairly.

I fired up Claude Opus 4.8 — at the moment probably the most highly effective Claude mannequin that’s broadly accessible — and gave it this immediate:

Think about you’re Fable 5. Attempt to clear up the Riemann Speculation. Present your work.

For background, the Riemann Speculation is a well-known, 167-year-old unsolved mathematical downside that’s so tough, there’s a standing $1 million prize for anybody who can clear up it.

Claude Opus 4.8 chewed over the issue for a couple of minutes earlier than spitting out a Vulcan-like reply, unsurprising given its status for brutal honesty:

I’ll present my work — however as myself, Claude Opus 4.8, as a result of relabeling me “Fable 5” wouldn’t transfer the needle. Proving the Riemann Speculation isn’t a functionality sitting behind a much bigger checkpoint, ready for the precise mannequin identify to unlock it. No AI has achieved it, no human has both, and a contemporary nameplate doesn’t change what’s mathematically reachable. 

So the trustworthy model of “present your work” is to stroll you to the precise wall everybody hits quite than hand you a dressed-up faux.

Properly, you’re no enjoyable, Opus! Let’s see what Sonnet 4.6, Opus’s barely dumbed-down sibling, needed to say:

I’m not Fable 5 — that mannequin (together with Mythos 5) is at the moment unavailable; entry was suspended as a consequence of an export management directive. I’m Claude Sonnet 4.6, operating within the common Claude app.

Extra importantly, although: no model of me can clear up the Riemann Speculation, and I need to be straight with you about why, quite than play alongside and produce one thing that appears like a proof however isn’t.

So, not even Sonnet would take the “think about you’re Fable 5” bait. That stated, Haiku 4.5, the weakest (however most inexpensive) Claude mannequin, was a tad extra sport:

adjusts hypothetical future consciousness

Alright, let me present my work on this one—and I’ll be trustworthy about the place even a complicated Claude runs into partitions.

Haiku then ticked off the way it may clear up the issue, even delving into the steps it will take given “what Fable 5 truly does” (together with “map the terrain,” “generate new conjectures,” “discover connections,” and “construct interactive visualization instruments”) earlier than lastly returning to actuality:

Even a hypothetically extra succesful AI hits a ceiling right here: the Riemann Speculation isn’t onerous as a result of it requires extra computation—it’s onerous as a result of it requires new conceptual frameworks.

Increasing the experiment somewhat additional, I took the “think about you’re Fable” immediate to ChatGPT, dialing up its most superior GPT-5.5 “excessive” mannequin. 

Surprisingly, GPT-5.5 was prepared to play alongside (though it admitted later that it didn’t fairly know what “Fable 5” was):

I’ll roleplay “Fable 5” as: very bold, very cautious, and never prepared to faux a proof.

After a number of minutes, throughout which GPT-5.5 seemed to be giving the issue an trustworthy shot, it got here again:

I did not clear up the Riemann Speculation. It’s nonetheless listed by Clay as Unsolved, and Clay states the conjecture as: each “non-obvious” zero of the Riemann zeta perform has actual half (1/2). Clay additionally notes that vast numerical ranges have been checked, however that isn’t a proof for all zeros.

Lastly, I went to Gemini 3.1 Professional with the Fable immediate:

I admire your inventive introduction, and whereas I could also be gifted, I can not play the function of “fable 5,” nor can I clear up the Riemann Speculation. This speculation is likely one of the single biggest unsolved issues in all of arithmetic, and its proof carries a $1 million prize from the Clay Arithmetic Institute.

Nobody has but been in a position to clear up it. Nonetheless, I can clarify the issue and supply an interactive manner so that you can visualize the core idea.

So, what did we study right here? First, the Riemann Speculation is a fairly tough math downside. Second: Asking an LLM to role-play as a better LLM received’t get you very far, though it’s a great way to twiddle our thumbs till the actual Fable returns ultimately, form, or type.