Amazon CEO's Reports Trigger U.S. Ban on Foreign Access to Anthropic Models

Amazon's Security Reports Led to U.S. Government Intervention

The Trump administration has halted all foreign use of Anthropic's most capable AI models, including Fable 5 and Mythos, following reports from Amazon CEO Andy Jassy. Jassy informed U.S. officials, including Treasury Secretary Scott Bessent, that Amazon researchers had successfully used specific prompts to elicit information from the Fable 5 model that could aid in executing cyberattacks—information that was intended to be off-limits.

Following these conversations, White House officials and security researchers tested Amazon's claims. The administration concluded that the most effective way to mitigate the risk was to prevent foreign governments, companies, and individuals from accessing the tools. President Trump subsequently signed off on the action for security reasons.

Anthropic's Response and Operational Impact

In compliance with the government's mandate, Anthropic shut off access to the Mythos and Fable models for all users. This decision has created several immediate operational challenges:

  • Global Research Hindrance: The shutdown potentially hampers global efforts to use these tools to identify and patch software vulnerabilities.
  • Internal Workforce Disruption: Because many of Anthropic's own researchers are foreign-born, the government's rule effectively prevented them from working on the latest models.
  • Competitive Risk: Analysts suggest that shutting down top models could hinder Anthropic's momentum as it prepares for a potential initial public offering (IPO) this fall, potentially driving users toward competitors like OpenAI.

Technical Debate Over Model Safety

There is significant disagreement between Anthropic and the reporting parties regarding the severity of the vulnerabilities discovered by Amazon.

Anthropic's Position

Anthropic maintains that the vulnerabilities flagged by Amazon are "relatively basic" and that other publicly available models are capable of discovering similar information. The company asserts that these findings do not represent a full "jailbreak" and that it has adequate safeguards in place.

Independent Expert Analysis

Andrew Morris, founder of GreyNoise Intelligence, noted that while Fable 5 could identify security bugs in at least four software programs when prompted specifically, this is "a long way from dangerous cybersecurity information." He highlighted that while Anthropic's software is known for its ability to convert bug information into working "exploit" code, there is no evidence that Amazon's researchers accessed that specific capability, which remains protected by security guardrails.

Political and Strategic Context

The crackdown is viewed by some as part of a broader escalation of government control over the AI industry. This move follows a recent executive order increasing security official oversight of AI models and discussions regarding the government taking equity stakes in AI companies.

Tensions Between Anthropic and the Administration

The relationship between Anthropic and the Trump administration has been characterized by long-standing friction:

  • Political Friction: The administration has expressed distrust of Anthropic due to its ties to liberal donors and the public criticisms of the administration by Anthropic CEO Dario Amodei.
  • Military Designation: The Pentagon previously designated Anthropic as a security risk, a move the company is currently fighting in two separate lawsuits.

The Amazon Paradox

Industry observers have noted the complexity of Amazon's role, as the company is a major investor in Anthropic and provides the chips for its data centers. While some view Jassy's actions as a genuine security warning, others speculate on the motivations of a major shareholder triggering a regulatory crackdown on its own investment.

Community Perspectives and Industry Implications

Discussion among technical communities suggests a divide in how this event is perceived:

"I still am struggling to understand why they informed the government about something that is known to be an issue in every LLM. There is no LLM that cannot be jailbroken... what was the reason to tell the government now?"

Other contributors pointed to the potential for this to set a dangerous precedent for open-source models:

"Unfortunately even if this blocking is only temporary, a precedent has been set. The government will likely be more willing to target open source models in the future that they deem to be too powerful."

Additionally, some users reported that Fable 5 provided a significant leap in productivity over competitors like GPT-5.5, suggesting that the loss of access to the model represents a tangible loss in technical capability for developers.

Sources