by Tan Aik Keong (AK)
AI never runs short of new stories, but the appearance and disappearance of Claude Fable 5 may be one of the most dramatic AI events of 2026.
On June 9, Anthropic officially released Claude Fable 5, billed as the company's most capable public AI system to date, and the first "Mythos-class" model opened up to the general public. Across software development, complex reasoning, scientific research and long-horizon autonomous task execution, it was described as far ahead of the previous Claude Opus line.
What came next was unexpected: just days later, the US government ordered a halt to the use of both Claude Fable 5 and Claude Mythos 5. By publicly available accounts, from its June 9 launch to being banned on June 13, it was public for roughly four days.
So what actually is Fable 5, and how does it differ from Mythos 5?
Start with what the two names mean. "Fable" comes from the Latin fabula — a fable, a story, a tale. "Mythos" comes from the Greek, meaning myth, epic, or a grand narrative about the world. Anthropic itself has noted the two words are etymologically close, both meaning roughly "a story that's told" — but the company deliberately used different names to distinguish two models with different levels of openness.
Technically, Claude Fable 5 and Claude Mythos 5 are near enough the same model. They share the same core architecture, parameter scale and reasoning capability. The real difference is in the guardrails.
Fable 5 is a tightly restricted public-facing version. When a user asks something touching on cyberattacks, bioengineering, chemical weapons or other high-risk research, the system automatically refuses, or switches to a less capable model to handle it.
Mythos 5 retains its full capability. Because of how strong it is at finding software vulnerabilities, analysing system weaknesses and complex scientific reasoning, it's only available to a small number of vetted institutions and researchers.
In other words, Fable 5 is the mythic model wearing a safety harness; Mythos 5 is the "complete myth" in the fullest sense.
In fact, Anthropic had spent months declining to release the Mythos line publicly, precisely because of concern about how strong its cyber-offensive capability was — the company had even said publicly that the model might be capable of discovering unknown software vulnerabilities.
Those safety concerns turned out to be well founded. According to reports, Amazon discovered a way to bypass Fable 5's safety mechanisms — a jailbreak — and the finding was passed to the US government. The White House and relevant regulators concluded that if those restrictions were broken, the model could be used to discover vulnerabilities in critical infrastructure, posing a national security risk.
On June 13, the US government issued an export-control order requiring Anthropic to stop offering Fable 5 and Mythos 5 to foreign users. Since there was no fast technical way to distinguish every restricted user from every other, Anthropic ultimately chose to pull both models entirely.
The episode points to a deeper problem: the pace of AI capability is starting to outrun the pace regulators can respond at.
We used to worry AI wasn't smart enough. Now we're starting to worry it's too smart.
Claude Fable 5's fate may be a preview of the new normal for future frontier models — where a technical breakthrough is no longer just an engineering challenge, but a complicated contest tangled up with national security, geopolitics and industry competition.
From Mythos to Fable, Anthropic had hoped guardrails would let its most powerful AI reach the public safely. What actually happened suggests that once a technology gets close to mythic capability, even the fable version of it may not be safe to let circulate. Editor's note: based on public reporting available at the time, Claude Fable 5 launched on June 9, 2026 and was taken offline or suspended on June 13 following a US national security and export-control order, putting its public availability at roughly four days.
Part of the AK AI Corner column. Originally published in Oriental Daily (东方日报) on Jun 14, 2026.