Anthropic Ban Blinds U.S. Defenders?

Washington just slammed the brakes on a powerful frontier AI tool, and the move has split the national security world in plain sight.

Quick Take

The White House ordered Anthropic to cut off foreign national access to Fable 5 and Mythos 5.
The order applies even to foreign national employees inside the United States.
Anthropic says the government acted after a narrow jailbreak report, not a proven model-wide failure.
Critics warn the ban may weaken American defenders more than it hurts adversaries.

What the Order Did

The Commerce Department ordered Anthropic to suspend access to Fable 5 and Mythos 5 for every foreign national, according to Anthropic’s public statement. That includes foreign nationals inside the United States and even foreign national employees at the company.^[1] The company said it disabled access for all users to comply while it challenges the order and tries to get the models back online.

The timing matters. Forbes reported the directive came just three days after Anthropic unveiled the models, which shows how fast the administration moved once the concern surfaced.^[6] Anthropic said the government relied on a reported jailbreak method that could bypass safety guardrails, but the company argued the issue was narrow and did not justify pulling commercial access from everyone.

Why Security Officials Cared

Federal officials acted on a national security concern tied to cyber abuse. Anthropic said the government received a report about an exploit that could help the model identify security flaws in critical systems.^[16] That is the core fear with frontier AI: a tool built for defense can also help attackers search for weak spots, write better prompts, and move faster than human analysts.

Anthropic’s defenders say the reported flaw does not prove the model is a unique threat. One analysis said the technique looked closer to a limited prompt trick than a full jailbreak, and Anthropic argued the same standard could stall almost every new model release.^[16] That is the central fight: whether the government saw a real emergency or overreacted to a contained risk.

The Split Inside Government

The sharpest irony is that the National Security Agency has reportedly been using Anthropic’s Mythos Preview for vulnerability scanning, even as the Pentagon labeled the company a supply chain risk.^[3]^[5] Reports said Anthropic limited Mythos access to about 40 organizations because it is strong at offensive cyber work, while those same reports also said the National Security Agency used it to find and fix weaknesses in its own systems.^[3]^[5]

That split gives critics an easy argument. If the federal government sees value in the model for defense, they ask why cut it off so broadly now.^[3]^[5] Cybersecurity Dive reported that dozens of experts said the ban could hand a gift to attackers by taking one of the best defensive tools away from U.S. teams.^[5] Supporters of the order counter that foreign access raises the stakes and that the government is trying to block reverse engineering before rivals exploit the model.

What Comes Next

For now, the models are offline for foreign nationals, and the dispute is headed into the wider fight over export controls and AI power. Forbes described the move as one of the most assertive uses of export-control authority against a commercial AI system.^[6] That description fits the mood in Washington, where national security officials want tighter control while industry leaders warn that blunt limits can slow innovation and weaken American cyber strength.

The bigger issue is trust. The public has not seen a full technical dossier explaining the government’s case, and that leaves room for doubt.^[2]^[16] In a country already tired of opaque federal power, this kind of order feeds a familiar concern: officials claim national security, but ordinary Americans are left to take it on faith while the real facts stay hidden.