Claude Fable 5被禁下架！白宮顧問揭露禁用內幕，楊立昆批自食惡果

Claude Fable 5 and Mythos 5 Disabled! A sudden U.S. export ban has triggered an uproar that forced them to be taken down. White House adviser David Sacks disclosed the inside story, pointing out that the company refused to patch a critical vulnerability—an action feared to deal a serious blow to its IPO plans this fall.

Claude Fable 5 and Mythos 5 Banned from Export, Forced Offline

A bombshell hit the U.S. AI industry last week. Because Anthropic’s latest models, Claude Fable 5 and Mythos 5, were unexpectedly subjected to an export control restriction by the U.S. government, they were forced to temporarily suspend service worldwide and be taken down—sparking a major uproar among the U.S. public and the tech community.

As a security warning report submitted by Amazon was revealed, and viewpoints from multiple senior political figures and prominent experts were published one after another, the complex security disputes and decision-making behind this ban—estimated to affect about 18.9 million monthly active users—are gradually coming to the surface.

Fable 5 Ban Reason: Did Amazon’s warning trigger a White House alert?

Last Friday, Anthropic suddenly issued a press release stating that, citing national security, the U.S. government had officially released an export control directive requiring Anthropic to fully suspend all access permissions for foreign persons to the Fable 5 and Mythos 5 models.

The scope of this ban is extremely broad, even including foreign employees working within the United States. To ensure compliance, Anthropic immediately decided to temporarily shut down all users’ access.

The key turning point driving this ban is closely related to Amazon. The Wall Street Journal reported that after Amazon CEO Andy Jassy’s researchers found a way to bypass Fable 5’s security protection mechanisms, he immediately contacted senior officials in the U.S. government—warning that the vulnerability could enable the model’s outputs to include sensitive information usable for cyberattacks.

Image source: Anthropic. Anthropic only just released Claude Fable 5 and Mythos 5 last week, but they were forced offline due to U.S. government export restrictions.

This warning triggered a high level of alert inside the White House. Although Anthropic later emphasized that the vulnerability was limited in scope and that other public models also had similar capabilities, tests by the UK AI Safety Institute also showed that the model could successfully bypass system defenses in as many as 73% of cases—prompting authorities to ultimately decide to take tough action.

White House Adviser: Anthropic Refused to Fix the Vulnerability, So Authorities Had No Choice

In response to this controversy, David Sacks, co-chair of the U.S. White House Council on Science and Technology, publicly stated that Claude Fable is essentially a Mythos model with added safety guardrails. Once those guardrails fail, its advanced cyberattack capabilities are exposed.

After receiving security warnings from trusted partner parties, the U.S. government asked Anthropic CEO Dario Amodei to fix the vulnerability or proactively take the models down, but it was explicitly refused.

Sacks revealed that the government was very surprised that Anthropic was unwilling to cooperate with this reasonable safety request—so the authorities, too, had no choice but to issue an export control directive.

He also criticized Anthropic’s public-relations effort to downplay the severity of the vulnerability, saying it runs completely counter to the company’s long-standing corporate spirit of putting safety first. The government’s position is very clear: as long as Anthropic is willing to properly resolve the security issues, the authorities are very hopeful of lifting the ban as soon as possible and getting Fable 5 and Mythos 5 back online.

Yann LeCun: Anthropic Reaps What It Sows

Beyond Sacks, Yann LeCun, the 2018 Turing Award winner and a well-known French scientist, also spoke out bluntly to criticize Dario Amodei.

He pointed out that Amodei’s previous absurd fear-mongering about the AI field and the Mythos and Fable models has finally brought about real consequences. The U.S. government has ultimately banned non-Americans from using the model, and even foreign employees inside the U.S. are restricted—everything is the result of Anthropic reaping what it sows.

In fact, David Sacks had already issued a warning as far back as 8 months ago, stating that Anthropic was carrying out a complex “regulatory hostage strategy” based on spreading fear, and that this kind of overemphasis on risk severely undermines the startup ecosystem.

Related report:
Anthropic Co-Creator Issues Warning: 50% Chance of Software Singularity Arriving—Could White-Collar and Blue-Collar Workers Both Be Impacted?

Anthropic CEO Warns: Mythos Has Found Tens of Thousands of Software Vulnerabilities—Only 6-12 Months Left to Patch

Coincidentally, on June 11, Anthropic’s CEO published a proposal on how the government should handle policy when AI development is moving too fast and regulation can’t keep up, which sparked heated discussion in the community.

Anthropic IPO Path Cast in Shadow

Previously, U.S. Defense Secretary Pete Hegseth labeled Anthropic as a supply-chain risk and triggered a lawsuit filed by Anthropic against the Department of Defense. Combined with this sudden issuance of a ban, it dealt an even heavier blow to Anthropic.

• Related report: Anthropic Sues the U.S. Government! Department of Defense Blacklist Case Opens—The Core Controversy in One View

At the same time, just last month Anthropic had secretly submitted an IPO listing application, originally expecting that it might be able to list on U.S. stocks this fall. But now, with the U.S. suddenly issuing a ban, its latest core products have been forced offline. Under the dual pressure of government oversight and market competition, Anthropic’s IPO path and global business layout are undoubtedly shrouded by a massive cloud of uncertainty.

Further reading:
Even the OpenAI and Anthropic CEOs Are Too Annoying! Doom talk and feelings of relative deprivation have made Americans turn against AI

OpenAI Considering Major Price Cuts! If it tries to win customers against Anthropic, why could it become the first shot in the AI bubble?