What happened at the White House in the 24 hours before Fable5 was forcibly taken offline?

Title: Inside the 24-Hour Whirlwind That Led the White House to Impose Export Controls on Anthropic

Author: Sophia Cai and Cheyenne Haslett, Politico

Translation: Peggy

Editor's Note: Anthropic's recently released Fable 5 model was suddenly taken offline just days after its launch. The reason was that the Trump administration, citing national security concerns, demanded restrictions on foreign citizens' access to two advanced AI models under Anthropic: Fable 5 and Mythos 5. Due to difficulties in precisely distinguishing user identities and access permissions, Anthropic ultimately chose to disable these models for all customers.

Behind this sudden takedown was a high-pressure game of negotiation between the White House and Anthropic within 24 hours. Discussions centered on whether the safety guardrails for Fable 5 could be bypassed. Anthropic CEO Dario Amodei engaged in multiple tense calls with government officials from the Treasury Department, Commerce Department, and White House cybersecurity leaders. The government believed the models could be used to identify software vulnerabilities, posing a national security risk; Anthropic emphasized that these issues were not "general jailbreaks" and criticized the government for acting without transparency, clarity, or a fact-based legal process.

The key issue in this incident is not just a model going offline, nor a breakdown in communication between the White House and an AI company, but the redefinition of advanced AI models as potential national security assets. When model capabilities extend into sensitive areas like cybersecurity, vulnerability discovery, and intelligence analysis, the government no longer relies solely on corporate self-assessment and voluntary review. Instead, it begins to use more forceful policy tools like export controls to directly influence the release schedule of these models.

Ironically, Anthropic has long been one of the most active advocates for regulation within the AI industry. Now, it finds itself the target of regulatory crackdown. The government considers its response to security vulnerabilities insufficient, while Anthropic argues that the White House overreacted. Behind this dispute lies a new challenge the AI industry must face: who has the authority to judge whether a model is sufficiently safe? Are corporate self-certifications enough? How will rapid government intervention under the guise of national security alter compliance costs, release timelines, and the global competitive landscape for AI companies?

The Anthropic incident may mark a watershed moment for the regulation of advanced AI. In the future, the launch of cutting-edge models will no longer be just a product release but a high-stakes deployment involving national security, geopolitics, capital markets, and industry order.

Below is the original text:

Before the Trump administration imposed comprehensive export controls on Anthropic, senior White House officials attempted a 24-hour emergency mediation to persuade the AI company to voluntarily take down a newly released model. Several officials believed the model posed a national security risk.

According to two government officials and a senior White House official, during this turmoil, Anthropic CEO Dario Amodei engaged in multiple tense phone calls with senior government officials, including U.S. Treasury Secretary Scott Bessent and White House cybersecurity chief Sean Cairncross. Due to the sensitive nature of the discussions, all parties requested anonymity.

This incident also brought to the forefront an increasingly urgent question: in the face of rapidly iterating, capability-expanding advanced AI models, how should the White House draw the line between innovation and safety?

These details of the calls had not been reported before.

Night Before the Model Was Taken Down: Safety Controversy

Following the implementation of export controls, Anthropic was forced to take down its new model Fable. The model had been publicly released only days earlier. Anthropic had previously assured the public that the model was safe and controllable, but shortly after release, senior government officials began reassessing whether its safety guardrails were truly as robust as claimed.

Two government officials and a senior White House official said that on Thursday, two days after Fable’s public launch, Amazon CEO Andy Jassy expressed concerns to the White House, suggesting that the model’s safety guardrails might be bypassed.

An individual familiar with Amazon’s communications said that Amazon was responding to government requests and providing feedback on potential risks.

By Friday morning, the issue had escalated to the highest levels of the White House.

According to officials, Bessent, Cairncross, White House Chief of Staff Susie Wiles, and other senior officials held a meeting to discuss the model and the government’s response. One insider said Bessent was en route to Houston for a scheduled public event and participated remotely.

After the meeting, the government attempted to contact Amodei but was told he was temporarily unavailable due to participation in a health retreat. Both a government official and a White House official confirmed this.

An Anthropic spokesperson denied the claim, stating: “This is completely false.”

An individual close to Anthropic said that the government first requested contact with Amodei around noon, and he was able to connect with senior officials within 1 hour and 15 minutes. The person added that during Amodei’s unavailability, Anthropic offered other executives to communicate on his behalf.

Once the government reached Amodei, he participated in three calls. A White House senior official and a government official said that about half a dozen senior officials took part, including Cairncross, Bessent, and Commerce Secretary Howard Lutnick.

The White House official said other White House staff and government officials also joined parts of the calls, including Jeffrey Kessler, Deputy Secretary of Commerce for Industry and Security; White House Secretary Will Scharf; White House Deputy Chief of Staff Richard Walters; and Presidential Policy Advisor Walker Barrett.

During the calls, Amodei sought to clarify what he believed were misunderstandings. He rebutted the government’s safety concerns, defended Anthropic’s protective mechanisms, and emphasized that the bypass methods observed were limited to specific scenarios and did not constitute a broad “jailbreak”—i.e., completely freeing the model from Anthropic’s safety guardrails.

In a blog post published after the export controls were enacted, Anthropic stated: “No testers have yet found a universal jailbreak—meaning a method to broadly bypass model safety measures and unlock extensive web capabilities.” The company also said that it is currently unrealistic for Anthropic or any other AI firm to completely eliminate all jailbreak attempts.

Anthropic also defended its safety system, claiming it is “so robust that many users complain it is overly broad.”

However, Cairncross and Bessent were not convinced by Amodei’s explanations. A White House official said that Amazon’s findings had been submitted for review to the NSA, which deemed the evidence sufficient to constitute “proof.”

According to White House officials and two government officials, the government urged Anthropic to voluntarily take down the models and coordinate with the government to fix vulnerabilities. Amodei requested more time and information but did not promise to remove the models. A White House official said that in one call, Bessent directly told Amodei he was making a “bad decision.”

Regulatory Iron Fist Falls: Fable 5 and Mythos 5 Under Control

Shortly after the calls, the Trump administration imposed export controls on Fable 5 and Mythos 5. Anthropic said the government invoked national security authority to prohibit foreign citizens from using these models. The company stated that the actual effect of the order was that, to ensure compliance, Anthropic had to “suddenly disable” access to these models for all customers.

A White House senior official said: “We pleaded with them for hours, and export controls were the last resort. It’s not what we wanted to do, but we had no choice.”

After the report was published, an individual close to Anthropic countered, saying the company was not given the so-called “voluntary cooperation” option.

This person said: “The White House only gave 90 minutes to take down the models, with no details about the actual threat. There were no pleas, no requests to cooperate—only a announced 90-minute deadline.”

A White House senior official said that White House officials had heard Amodei compare the danger of Anthropic’s technology to nuclear weapons. Therefore, when this CEO, who has long advocated AI safety regulation, expressed reluctance to take down the system to fix known security issues, officials found it hard to understand.

Anthropic has long been regarded as one of the most active companies advocating for AI regulation. The company has repeatedly argued that regulatory frameworks are needed to address the global security risks and employment impacts that rapid AI development could bring.

Three individuals familiar with government thinking said that Amazon was not the only company raising concerns with the government.

One said: “The core issue is that Anthropic has not taken this matter seriously enough. If Anthropic had treated it seriously, rather than dismissing it as an isolated incident, and actively fixed or paused access, this wouldn’t have escalated to this point.”

Another close to Anthropic countered that the so-called “jailbreak” does not mean Fable 5’s security system has failed, and pointed out that the company had maintained communication with the government before releasing Fable. The person said that during multiple rounds of communication, the government did not oppose the release of Fable.

After the export controls were enacted, Anthropic stated in a blog that it would comply with government directives but also believed the move was excessive.

Anthropic said: “As we have previously stated publicly, we believe the government has the right to block unsafe deployments, but this should be based on a transparent, fair, clear, and technically fact-based legal process. This action does not meet those principles.”

An anonymous White House official said that innovation remains the White House’s “top priority,” but safety must also be prioritized.

Amazon declined to disclose specific details of its communications with the government. A spokesperson said: “It’s not uncommon for the government to seek our input on potential security risks. When that happens, we do not disclose details of those discussions.”

In early April, Anthropic announced that its latest powerful model Mythos would be available only to a select few tech and cybersecurity companies to help test software vulnerabilities. The company said that because the model’s capabilities are so strong, if it fell into the wrong hands, it could cause serious consequences, so its release was limited.

The debut of Mythos triggered a series of meetings between Amodei and senior White House officials. Both sides described these meetings as productive. Subsequently, they engaged in multiple discussions on how to regulate advanced AI models, ultimately leading to a recent executive order requiring companies to voluntarily submit models for government review before widespread deployment.

The publicly released Fable 5 this week was described by Anthropic as a “Mythos-level model” but with additional safety measures enabling it to be accessible to general users. The model had been reviewed by the U.S. government and the UK’s AI Safety Institute.

However, after the so-called security flaw was disclosed, several government officials believed the model should be immediately taken down.

On Saturday morning, former White House AI chief and staunch anti-regulation advocate David Sacks posted on X, supporting government action against Anthropic.

Sacks said he did not believe the jailbreak was trivial or irrelevant, nor did he think export controls were an attempt by the government to exert broader control over the AI industry. He wrote: “The government now wants Anthropic to fix the security issues, then lift the export controls, and let Fable go back to public release. They want all this to happen quickly. Frankly, what’s confusing is that Anthropic has repeatedly said that security is its top priority, yet now they’re unwilling to cooperate.”

Sacks and other officials had previously criticized Anthropic multiple times, accusing it of left-wing political bias and warning about large-scale employment impacts, claiming the company was creating panic.

Earlier this year, Defense Secretary Pete Hegseth and the Pentagon escalated the dispute with Anthropic to unprecedented levels. On March 3, the Pentagon listed the company as a supply chain risk because Anthropic refused to allow its AI tools to be used for mass domestic surveillance and autonomous weapons systems.

On Saturday, Sacks said that the ongoing dispute between the government and Anthropic was unrelated to the recent export controls. He wrote: “The government values Anthropic’s technical capabilities and considers this issue serious but easily solvable. The ball is now in Anthropic’s court.”