Claude Fable 5 was taken offline globally just four days after its release. Developer Jamieson O'Reilly used a leaked 120,000-character system prompt to successfully "resurrect" it and inject it into the live Opus 4.8 model with a single line of code. Experimental results showed that merely replacing the system prompt could make the same model exhibit entirely different "personalities" and output styles. According to Information, the trigger for the takedown was Amazon, Anthropic’s primary investor, which obtained sensitive information through jailbreak prompts during internal testing and immediately reported it to the White House. Anthropic was given only 90 minutes to respond; CEO Dario Amodei refused to patch the vulnerability, leading to Fable 5’s forced shutdown. This incident has for the first time brought to light the paradox of the ASI era: when safeguards are breached, model capabilities become dangerous weapons, and prompt-based constraints appear increasingly fragile.
Article author and source: AI New Era
Claude Fable 5, once declared "dead," has come back to life!
Just now, developer Jamieson O'Reilly used a leaked system-level prompt to forcefully unlock the lightweight version of Fable 5.
A single line of code injection was enough to instantly "awaken" Opus 4.8.
In the extreme control experiment, O'Reilly was given the same instruction—to create a modern Apple-style webpage.
The left screen, infused with the soul of Fable 5, outputs a web page that feels entirely like a different "personality."
Now, calls for Claude Fable 5's return are growing louder across the entire network.
Everyone planned to make a big move over the weekend with Fable 5, but upon waking up, they found their internet disconnected by Anthropic.
The Fable 5 Build conference is being held this weekend; the temporary downtime was truly unexpected.
Meanwhile, the deeper内幕 behind the sudden ban of Fable 5 has finally come to light.
Information leaks, which prompted the U.S. government to issue a ban, were directly linked to Anthropic's primary investor—Amazon.
What exactly is going on here?
One line of code brought Fable 5 back to life.
However, before delving into this shocking revelation, let’s first address everyone’s most urgent question:
What kind of miraculous move is needed to bring Fable 5 back to life?
Just 24 hours after the birth of Fable 5, the top jailbreaker, Pliny the Liberator, released a major file on GitHub:
Claude Fable 5——System Prompt
A full 120,000 characters, 1,585 lines, 72 named sections, and even the JSON definitions for all 18 tools have been extracted.
Address: https://github.com/elder-plinius/CL4R1T4S/blob/main/ANTHROPIC/CLAUDE-FABLE-5.md
With it, you’ve obtained the core “personality blueprint” of Fable 5.
Unexpectedly, after Fable 5 was taken offline, this document became the only "lifeline" for players across the internet to bring it back to life.
This led to the opening scene, where developer Jamieson stared at the leaked prompt and conceived a bold idea—
Since Fable 5 has been delisted, I want to figure out—
How much of its "personality" is shaped by system prompts, and how much is determined by the model itself?
So he got started—the process was absurdly simple: open Claude Code and enter a single command:
- claude --dangerously-skip-permissions --system-prompt-file CLAUDE-FABLE-5.md
Note the parameter: --dangerously-skip-permissions, which dangerously skips all permission confirmations.
This is a high-risk setting flagged with a warning by the official Claude Code: a green light that no longer prompts you with “Are you sure?”
He used this command to forcibly inject the leaked Fable 5 prompt into the active Opus 4.8.
Next, he opened another window on the right, running the native, unmodified Opus 4.8.
The same brain, the same intelligence, the same context. The only variable is that system prompt.
Then, he gave both models the exact same command: create a landing page in modern Apple style.
The results are in, and the difference is obvious.
Jamieson's exact words were, "The same intelligence produces things that are completely different species."
Simply changing the system prompt has caused the two web pages to diverge completely in brand tone, copy style, section structure, and overall character.
One is a straightforward, generic template; the other carries the meticulously refined aesthetics and expressive rhythm of Fable 5—
The layout, wording, and module breakdown all bear the imprint of the "factory personality."
This, even Anthropic would have to stay silent!
By the way, today’s launch of OpenRouter’s Fusion API achieved Fable 5-level intelligence at just half the price.
The big investor strikes back, triggering a shocking betrayal in the tech world
Next comes the real highlight.
Why was the highly anticipated ASI model forcibly taken down just four days after its release?
An exposé by foreign media has definitively unmasked the "mastermind" behind it all—Amazon.
During internal testing, the Amazon team used a carefully crafted series of prompts to bypass Fable 5’s safety safeguards, causing it to output sensitive information that should have been locked down and inaccessible for cyberattacks.
After receiving this result, CEO Andy Jassy did not privately reach out to Anthropic; he called Washington directly—
The people who answered the call included senior officials such as Treasury Secretary Scott Bessent.
Keep in mind that Amazon is a major shareholder in Anthropic, supplies it with data center chips, and is deeply integrated with it—more so than ever.
The so-called "big spender" then handed the test results directly onto the White House desk.
90 minutes, global internet shutdown
The White House responded with astonishing speed; officials convened urgently, and security experts worked through the night to replicate Amazon's test results.
They then issued Anthropic an ultimatum: either fix the vulnerability or take down the model.
At the time, the remaining window was as short as 90 minutes!
Anthropic stated in its statement that the letter did not specify what the national security concerns were, so they had no choice but to comply and shut down Fable 5.
This cut hurts the entire world. Countless companies that rely on these tools to identify software vulnerabilities are suddenly left without resources.
More absurdly, a large number of foreign researchers at Anthropic, due to this order, cannot even access their own latest model and are unable to do their work.
Karpathy is one of the banned employees.
Dario stands firm: I refuse, tearing up the "get-out-of-jail-free card" myself
The truly dramatic moment came when renowned investor David Sacks published a lengthy article.
He opened by saying the most fatal thing: Fable 5 is essentially Mythos with safety guardrails.
Mythos possesses powerful advanced cyberattack capabilities, and the purpose of the safeguards is to lock away these "weapons."
Once the safeguard fails, this ability is exposed to those who shouldn't have it.
The key point is that a highly trusted partner, during testing of Fable 5, presented an escape method capable of breaking through the护栏.
At the time, the government required Dario to fix the vulnerability or take down the model directly—
Dario declined!
In this lengthy article, Sacks goes all out: Anthropic downplays the vulnerability as "not serious" on its blog while continuing to operate consumer-grade models.
In short, they put business ahead of security.
This is completely inconsistent with its claimed identity of “security first” and “AI company focused on security.”
It is understood that the government is very reluctant to resort to bans and is surprised that Anthropic refused to cooperate even with a reasonable request to fix a vulnerability.
Admittedly, this narrative is extremely powerful.
But in Anthropic’s official blog response, they used just one word to characterize it: “misunderstanding.”
What’s most heartbreaking is that they also dragged GPT-5.5 into it, saying exactly this—
The same technique can be replicated on other open models, including OpenAI's GPT-5.5.
Fable 5 tore open that gap in ASI.
What made this incident most memorable was that it first laid before everyone the sharpest paradox of the ASI era.
A model so powerful that its safeguards are breached is equivalent to loss of control. The balance between safety and capability has reached an unprecedented height.
Fable 5 is essentially just a wrapper around Mythos.
Once the shell cracks, the carefully locked abilities inside are exposed to those who shouldn’t have them.
And the closer it gets to superintelligence, the more this safeguard resembles a thin film—
Yet today, just a leaked system prompt and a single injection command are enough to bring it back to life.
Tomorrow, when the model’s capabilities become so powerful that even prompts can no longer contain them, how much reaction time will be left for humans?
Models can be taken down overnight, but Pandora's box has no such option as "temporarily unavailable."