US Crackdown Triggered by Amazon Warning on Anthropic AI Models

The Trump administration’s decision to cut foreign access to Anthropic’s most powerful AI models was reportedly set in motion by outreach from Amazon CEO Andy Jassy, after researchers found a potential way to coax the company’s “Fable 5” model into producing information that could be repurposed for cyberattacks. The episode escalated quickly into White House-level review and culminated in export controls that, according to Anthropic, followed a U.S. directive.

As the dispute unfolded, U.S. regulators also demonstrated how fast access to a widely used AI system can be curtailed—an event that traders linked to a rebound in several decentralized “AI token” projects on Friday and Saturday. The move has reignited questions about how AI safety issues are assessed, communicated, and ultimately enforced across borders.

Key takeaways

• According to The Wall Street Journal, Amazon CEO Andy Jassy contacted senior U.S. officials after Amazon researchers identified a prompt-based workaround affecting Anthropic’s Fable 5.
• The White House reportedly moved to impose export controls and engaged Anthropic leadership; Anthropic pushed back and said it was working to restore access for users.
• David Sacks, co-chair of the President’s Council of Advisors on Science and Technology, said the administration acted “reluctantly” and hoped Anthropic would fix the jailbreak issue quickly.
• On-chain and token-linked markets appeared to react to the sudden availability shift, with decentralized AI-related tokens posting gains over the same period.

How the alleged “jailbreak” concern reportedly triggered export controls

In a report cited by The Wall Street Journal, the core catalyst was described as a method discovered by Amazon researchers that could prompt Anthropic’s Fable 5 into returning information that might be used in cyberattacks. The WSJ reports that Jassy reached out to senior U.S. officials on Thursday after that finding, along with similar warnings from at least five other firms.

Those escalations reportedly led to a “frantic shuffle” within the White House to assess the threat and to contact Anthropic CEO Dario Amodei. According to the Politico reporting referenced in the original account, Amodei pushed back on the administration’s concerns and asked to resolve them through voluntary cooperation rather than immediate removal of access.

Anthropic ultimately said it believed the U.S. directive stemmed from a misunderstanding of the threat, describing the issue as a “non-universal jailbreak” that came from an unnamed report. In earlier coverage from Cointelegraph, Anthropic was reported to have suspended access to its new model—an action tied to the U.S. directive.

What Anthropic and U.S. officials said—and what remains disputed

One signal of where the disagreement may lie is the framing of the risk. Anthropic’s blog post, as described in the reporting, suggests the administration’s understanding of the jailbreak’s scope may have been overstated. The company’s view was that the jailbreak was not “universal,” implying it would not reliably work across prompts and contexts.

Still, U.S. officials acted decisively. David Sacks told X on Saturday that the administration issued the export control in response to the situation and that it was “very surprised” Anthropic did not want to cooperate on a “reasonable safety request” to fix the jailbreak issue. Sacks characterized the administration’s approach as cautious and reluctant, but nonetheless committed to the safety request and the goal of restoring access after remediation.

The original reporting also states that Amazon did not confirm whether it spoke directly with government officials about the models. In an emailed statement, an Amazon spokesperson said that it is common for governments to seek counsel on potential security risks and that the company does not share details of those discussions when they occur. That leaves some of the exact internal communications unclear, even as the WSJ account links Jassy’s outreach to the administration’s subsequent decision.

Why the “switch-off” matters for AI providers and users

Beyond the specific model dispute, the episode highlights a practical reality for AI users and integrators: access to a U.S.-linked model can be restricted quickly, even for high-demand systems. Cointelegraph’s earlier coverage noted that the directive forced Anthropic to pull its model from public availability, and the current reporting ties the export control to the same turning point.

For enterprises that depend on frontier models—whether for customer-facing applications, research workflows, or operational automation—the risk is not only technical but operational. A safety finding, a regulatory determination, or even a disagreement about the severity of a vulnerability can translate into abrupt availability changes. That can affect uptime, roadmap planning, compliance reporting, and model routing strategies across vendors.

Anthropic stated it is working to restore access for its users. Sacks’ comments add what many market participants will likely focus on next: whether remediation happens fast enough to satisfy regulators and lead to removal or easing of the export control.

Token markets react to sudden access restrictions

The crackdown also appeared to ripple into crypto markets tied to decentralized AI narratives. The original reporting describes the U.S. government’s ability to rapidly disable access to U.S.-based AI models as a factor behind gains in several decentralized AI tokens over Friday and Saturday.

According to the cited CoinGecko pricing data, Bittensor’s native token rose 23.9% over the past 24 hours referenced in the article. Bittensor is described as a decentralized AI protocol enabling people to build and monetize AI models. The reporting also cited Venice Token (VVV) up 16% and Near Protocol’s token rising 6.2%, with Near positioned as infrastructure for decentralized AI agent applications.

While these moves do not necessarily indicate a direct causal relationship between the export control and token valuations, the timing underscores how quickly crypto markets can respond to policy-driven changes in the broader AI ecosystem. For traders and long-term holders, the key question is whether the market is reacting to the news itself or to expectations that decentralized models and infrastructure will gain relative attention if centralized providers face recurring regulatory constraints.

As the situation develops, readers should watch for concrete updates from Anthropic on remediation efforts, as well as any indications that regulators are prepared to lift or modify the export controls. The central uncertainty remains whether the jailbreak risk is viewed as sufficiently resolved—and how regulators will define “enough” safety to re-enable broader access.