Amazon CEO Raises AI Model Concerns, Prompting Government Intervention and Global Block on Two Models

Amazon CEO Andy Jassy reported security concerns about language models developed by Anthropic, an AI company in which Amazon has invested, to senior U.S. government officials, according to multiple reports. This triggered the U.S. government to impose export restrictions on two models developed by Anthropic, leading the company to block access to these models worldwide.

The sequence of events highlights the complex relationship between AI developers, their key investors, and regulators. According to TechCrunch, the Wall Street Journal first reported the matter. Jassy reportedly told government officials, including Treasury Secretary Scott Bessent, that Amazon researchers had used Anthropic’s “Claude Fable 5” and obtained information that could be exploited for cyberattacks.

From Notification to Regulation and Blockade

Following Jassy’s information, the government imposed a ban under export controls on two models: Claude Fable 5 and “Mythos 5.” In response, Anthropic decided on Friday (June 12) to block global access to both models.

An Amazon spokesperson told TechCrunch that “it is not uncommon for the government to seek our views on potential security risks,” but declined to share details of these discussions. The spokesperson also noted that AWS had published an update indicating it was affected by the model block.

The Information and Reuters have also reported similar accounts, painting a picture where Amazon leveraged its position as Anthropic’s major investor to communicate model security concerns to the government.

Sacks’ Claims and Anthropic’s Rebuttal

David Sacks, who served as the AI czar under the Trump administration and now co-chairs the President’s Council of Advisors on Science and Technology, has offered his own explanation. According to Sacks, “a very trusted partner for both Anthropic and the U.S. government” discovered a jailbreak (a method to bypass security restrictions) and reported it to the authorities.

Sacks stated, “The administration asked Anthropic CEO Dario Amodei to either fix the jailbreak or stop deploying the model. Dario refused.” If true, this would mean Anthropic did not comply with the government’s request, leading to the forced access blockade.

Anthropic, however, argued in an official blog that the capabilities the government flagged were already available in other publicly released models. Anthropic’s position is that its models are not uniquely dangerous.

The Complex Amazon-Anthropic Relationship

Notably, Amazon is one of Anthropic’s largest investors, having poured billions of dollars in total into the company. Amazon Web Services (AWS) provides Anthropic’s primary computing infrastructure, indicating a close collaborative relationship.

The fact that Amazon’s CEO reported concerns about a product from a company his own firm invested in underscores the challenges of conflicts of interest and information management within a corporation. What judgment was made between the role as an investor and the responsibility to share findings from its own security research team with the government? The Amazon spokesperson explained that “it is not uncommon for the government to seek our views on security risks,” but in this case, it escalated into a major action—government export restrictions and a complete model blockade.

Impact of the Model Blockade

Anthropic’s global blockade of access to Claude Fable 5 and Mythos 5 directly affects its users. Companies and developers that used these models on AWS, in particular, must now seek alternatives.

The Amazon spokesperson acknowledged that AWS is affected by the model block, meaning customers using Anthropic models on the cloud service may face service interruptions or feature limitations. While Anthropic’s blog post notes that the capabilities in question are already available in other models, users still face migration costs and compatibility issues.

New Phase of AI Safety and Regulation

This case could set a new precedent for government-corporate relations concerning AI model safety. Previously, it was common for AI companies to voluntarily assess safety and sometimes refrain from releasing models. However, the government’s use of export controls based on information from an investor represents a substantive shift in the regulatory framework.

If David Sacks’ account—that “the administration asked for a fix or halt in deployment, but Anthropic refused”—is true, tensions between the government and AI companies could deepen. Anthropic has consistently positioned itself as a safety-focused company, and claims that it rejected a government request would send shockwaves through the industry.

Editorial Opinion

Short-term Impact

Over the next 3–6 months, the security evaluation process for AI models is likely to change significantly. This case demonstrates that vulnerabilities discovered by an investor’s research team can lead directly to regulation via the government. AI companies will have no choice but to tighten their own model security assessments. Additionally, major tech companies that are key AI investors are expected to demand more detailed security audits of the models they fund.

Long-term Perspective

Over a 1–3 year span, export controls on AI technology may become more common. Similar to semiconductors and military technology, specific AI models could increasingly be subject to export management as “dual-use technologies.” This could push open AI research communities toward a more closed-off direction, raising concerns about mercantilist hoarding of models. Meanwhile, Anthropic’s argument that “the same capabilities are available in other models” raises fundamental questions about the effectiveness of such regulations.

Editorial Questions

How should the act of an investor company’s CEO reporting concerns to the government about a product from a company they have invested in be evaluated from a corporate governance perspective? Furthermore, if the government imposes export controls based on internal information from a specific company, how will due process for other AI companies be ensured? There is also a need to discuss the fairness of regulating only some models when the same capabilities are widely available in others.

References

• TechCrunch: Amazon CEO reportedly raised Anthropic model concerns before government crackdown — Published 2026-06-13
• Wall Street Journal report (referenced in the original article)
• Related reports from The Information and Reuters (referenced in the original article)
• Anthropic official blog post (referenced in the original article)