Early this morning, according to multiple foreign media reports, Anthropic announced that it would terminate access to the Fable 5 and Mythos 5 models. This may be the trigger.Amazon CEO is Andy Jassy. The Wall Street Journal reports that while conducting research using Anthropic’s Claude Fable 5, Amazon researchers obtained information that could be used in cyberattacks. Subsequently, the U.S. government imposed an export control ban on Fable 5 and Mythos 5.

David Sacks, Trump’s former AI commissioner and current co-chairman of the President’s Council of Advisors on Science and Technology, posted on social platformAnthropic CEO Dario Amodei refuses to fix, and claimed in subsequent official announcementsThis vulnerability is not serious and can be found in models from other companies..

Sacks believes that this downplaying of risks is inconsistent with Anthropic’s brand positioning of itself as an AI security company. He also revealed that the U.S. government hopesAnthropic could resolve security issues quickly to lift export controls, and re-release the Fable 5 model.


To compensate for user rights, Anthropic providesRefund channel, some netizens posted the refund email they received from Claude. The email mentioned that Anthropic has closed access to the Claude Fable 5 model; all other models of Opus, Sonnet, and Haiku can be used normally; and the user's usage quota has been reset.

If the user feels that this adjustment does not meet the user's needs, they can choose to cancel the subscription and automatically receive a refund based on the remaining time. The refund deadline is June 20, local time. This refund does not support operations on the Claude mobile app.


▲Refund email received by netizens (Source: X)

1. Claude started refunding, netizens said: There is no turning back

Many netizens mentioned that Anthropic’s refund processing plan to ban Fable 5 hastily released has many bugs. One of the netizens said that Anthropic directly shut down all the user's subscription services. His previous subscription rights that were five times longer and valid until June 19 were gone, which was outrageous.


Some netizens directly asserted that Claude has been eliminated and there is no way back.



Some netizens analyzed that Anthropic’s refund operation also proves that Fable 5 will not come back soon.


Some netizens said they would cancel their subscription and wait for the new model of ChatGPT.


2. The U.S. government claimed that Anthropic avoided important issues and weakened security vulnerabilities in new models.

An Amazon spokesperson told The Information that as a leading cloud services provider serving a large number of private and public sector customers, it is not unusual for the U.S. government to seek Amazon's advice on potential security risks. Even if relevant communication is carried out, we will not disclose the specific content of the negotiation.

The spokesperson also revealed that Amazon Cloud Technology has also been affected by Anthropic's ban model. On the morning of June 13, Amazon Cloud Technology revoked access to Fable 5 and Mythos 5 in all regions.

The Information quoted people familiar with the matter as saying that after talks with Jassy and other technology company leaders, US National Cyber ​​Affairs Director Sean Cairncross convened a meeting of senior US government officials on Friday, at which the decision was made.Export controls are the most direct means of taking action against Anthropic.

After the meeting, Anthropic CEO Dario Amodei spoke with U.S. government officials. During the meeting, Anthropic explained that the so-called jailbreak vulnerability exposedThe threshold for safety hazards is not too high, other large models also have the possibility of similar problems. However, the U.S. government informed Anthropic that the decision to implement export controls has been finalized.

David Sacks, Trump’s former AI commissioner and current co-chairman of the President’s Council of Advisors on Science and Technology, posted some details on the social platform X.

He said that a reliable partner who is highly trusted by both Anthropic and the US government took the initiative to expose a model jailbreak vulnerability. And when the U.S. government asked Amodei to fix the jailbreak vulnerability or else the model would be removed from the market, Amodei refused to fix the vulnerability.

Sacks also mentioned that Anthropic claimed in its official blog that the jailbreak vulnerability was not serious. However, neither the reliable partner nor the U.S. government agreed with this statement. This statement of downplaying risks also contradicts Anthropic's brand positioning of being an AI security company.Currently, the U.S. government hopes Anthropic can resolve the security issues as soon as possible to lift export controls and re-release the Fable 5 model.

In addition, Anthropic also mentioned in the blog announcing the banned model that similar vulnerabilities have appeared in other public models. However, an official close to the U.S. government told The Information that the U.S. government is unlikely to extend export restrictions on Anthropic’s advanced models to other AI companies.

Amazon is a major investor in Anthropic. Amazon publicly mentioned in April this year that it has invested US$13 billion (approximately RMB 87.9 billion) in Anthropic since 2023, and plans to invest another US$20 billion (approximately RMB 135.3 billion) in the future. Amazon’s Trainium chips help train and drive Anthropic’s large models, which also run on Amazon’s cloud services.

It is worth mentioning that this year Amazon also made a large investment in Anthropic’s main competitor OpenAI, promising to invest up to US$50 billion in it, and for the first time it will connect OpenAI’s latest model to Amazon Cloud Technology for external sales.

Another major background of the U.S. government’s ban on the Anthropic model is that the U.S. government issued an executive order two weeks ago to launch a voluntary compliance mechanism: AI companies can submit models to the government for verification before releasing models to partners and the public. This is because there has been a debate within the U.S. government over whether to establish a licensing system with higher thresholds and more stringent review for AI companies that plan to launch high-end large models. This executive order is the product of this game.

Conclusion: Anthropic’s strongest model was shut down, and the industry’s safety threshold was forced to rise?

On June 10, Anthropic officially released two major models, Claude Fable 5 and Mythos 5. They both belong to the Mythos (myth) level and share the same underlying model. Their capabilities are higher than the Opus level. This is also Anthropic's most powerful and highest overall performance large model level to date.

During the same period of release, Anthropic deliberately designed hierarchical openness and built-in security downgrade routing. It originally wanted to balance strong capabilities and risk management and control, but it still failed to withstand the chain impact caused by hidden security vulnerabilities. In just a few days, two flagship models were shut down. This may mean that before new models are launched in the future, security testing related to in-depth attack and defense verification by a third-party independent red team may become a hard threshold.