The Government Pulled the Plug on Anthropic’s Newest AI Models

Hand drawing artificial intelligence digital circuit board.

Washington’s move against Anthropic’s Fable 5 and Mythos 5 shows how fast AI policy can turn into a national security crackdown that hits defenders, researchers, and foreign employees at once.

Quick Take

The United States Department of Commerce ordered Anthropic to cut off foreign national access to Fable 5 and Mythos 5.^[16]
Anthropic said the order applies to foreign nationals anywhere, including its own foreign employees.^[1]^[16]
Officials did not publicly spell out the full evidence behind the directive, which leaves the justification open to dispute.^[4]^[16]
Critics say the ban could weaken United States cybersecurity work by taking a useful tool away from defenders.^[5]^[7]

What the Order Did

On June 12, the Department of Commerce told Anthropic to suspend access to Fable 5 and Mythos 5 for foreign nationals, and Anthropic said it complied by turning off access for all customers while it challenged the order.^[16] Anthropic also said the directive covers foreign nationals whether they are inside or outside the United States, including foreign national employees.^[1] That broad reach makes this more than a simple sales limit.

The government’s stated reason was national security, but public reporting says officials did not release a detailed technical file explaining why the models had to be shut down.^[4]^[16] Reporters said the order followed a report about a jailbreak-style exploit that might let the model bypass safety guardrails and help find cybersecurity weaknesses.^[16] Anthropic disputed that read, saying the issue looked narrower than a full model escape.

Why Security Officials Moved So Fast

The speed matters because the directive came only days after Anthropic unveiled the models.^[6] That quick response suggests the White House saw a fast-moving risk, not a slow policy problem. Reported concerns centered on cyber misuse, not ordinary consumer harm. One report said Amazon researchers found a prompt-based jailbreak method in Fable 5, which raised alarms that the model could be pushed into dangerous security work.^[6]

Still, the evidence in public view is incomplete. A technical report showing the exact exploit has not been released, and that gap makes it hard to judge whether the threat was unusual or simply another example of a frontier model being tricked.^[4]^[16] Anthropic said the technique appeared limited and did not show a broad unlock of cyber powers. That leaves the core dispute unresolved for outside analysts.

Why Critics Say the Ban May Backfire

Opponents argue the ban may hurt the same people it is supposed to protect. Cybersecurity experts told reporters that the restriction takes strong models away from defenders who use them to find flaws before criminals do.^[5] Some also warned that a blanket cutoff could weaken United States leadership in advanced AI, especially if foreign competitors are not bound by the same limits.^[5]^[7] That argument cuts across party lines.

Reuters: A US legal tech company just sued the US federal government over the order of forced Anthropic's model shut down of Fable 5 and Mythos 5.

Legion LegalTech (a U.S.-based AI-native litigation-technology company) says the order immediately broke its workflow because its… pic.twitter.com/boGQlfwxv7

— Rohan Paul (@rohanpaul_ai) June 24, 2026

The wider context is a growing split inside the national security system. Separate reports say the National Security Agency has used Anthropic’s Mythos Preview for vulnerability scanning even as other parts of the government treated the company as a supply-chain risk.^[3]^[5] That tension helps explain the uproar. One branch of government appears to value the model as a shield, while another treats the same kind of tool as too risky to keep online.

What This Says About U.S. AI Policy

This case fits a larger pattern of emergency-style export controls that move faster than public explanation. It also shows how quickly AI policy now touches defense, trade, labor, and trust in government. For critics on the right, the episode looks like more federal overreach and weak transparency. For critics on the left, it raises questions about corporate power, unequal access, and whether national security claims are being used without enough proof.

The lasting issue is not just one model or one company. It is whether the government can draw a clear line between real danger and broad fear. If the line is too loose, useful tools disappear from the hands of defenders. If it is too weak, foreign actors may exploit systems faster than agencies can respond. That is the policy trap now sitting at the center of the AI fight.