Anthropic Sends Security Team to Washington. The Fable 5 Shutdown Is Now a Negotiation.

Since the export control order landed at Anthropic's offices at 5:21 p.m. ET on Friday, June 12, the standoff between the San Francisco AI lab and the Trump administration has shifted from a regulatory dispute into an active negotiation.

Who Is in the Room

According to The Wall Street Journal, Anthropic sent three of its top security researchers to Washington over the weekend: Nicholas Carlini, Logan Graham, and Dave Orr. They met with Commerce Secretary Howard Lutnick, National Cyber Director Sean Cairncross, and Treasury Secretary Scott Bessent. Sending a Commerce Secretary and a Treasury Secretary to talk about a jailbreak patch signals the administration treats this as a genuine national security matter, not bureaucratic posturing.

The Wall Street Journal also reported that people close to both parties say they want to resolve the issue and restore access to the models. But critically, nobody has described what a solution actually looks like.

What Anthropic Says Happened

In its June 12 public statement, Anthropic was precise and pointed. The government gave Anthropic verbal evidence of a potential jailbreak technique for Fable 5. When Anthropic reviewed a demonstration of that specific technique, the company says it found only a small number of previously known, minor vulnerabilities, all of which other publicly available models could also surface without any bypass at all.

Anthropics's position: its safeguards are the strongest ever deployed by any frontier lab, thousands of red-team hours were spent before launch with the U.S. government, the UK AISI, and private third parties, and no universal jailbreak has been found. The company acknowledged openly that perfect jailbreak resistance is not achievable by anyone in the industry today, and said it disclosed that limitation at launch.

Anthropic also pushed back directly on the order's proportionality, saying the standard applied here "could essentially halt" new frontier AI models if used broadly, and that the action "does not adhere" to principles of a transparent, fair, and technically grounded process.

What the Administration Says

David Sacks, the White House AI & Crypto Czar (Special Advisor for AI and Crypto), posted on X that the government warned Anthropic that Fable 5 could be jailbroken and that CEO Dario Amodei was notified but the company did nothing to fix it, according to Al Jazeera's reporting on June 14. Sacks framed it as Anthropic refusing to act, not Anthropic disagreeing about severity.

Anthropic's public statement does not dispute that it received the government's concern. It disputes whether the demonstrated technique warranted the scope of the response.

The China Access Question Is Still Open

As reported in our June 14 coverage, Semafor reported that a China-linked group may have accessed Mythos before the shutdown, and that the White House never informed Anthropic that this was a driving reason for the export order. As of June 15, that specific allegation remains unconfirmed and no investigation or charges have been announced publicly.

Al Jazeera noted that China has made significant strides in AI development, including DeepSeek's V3 and R1 models, and that Beijing controls critical rare earth supply chains feeding the sector. The strategic anxiety behind the order is real regardless of whether the specific China-access allegation is proven.

The Legitimate Concern Worth Taking Seriously

Critics of Anthropic's position have a reasonable argument: a model ranked first globally for intelligence, with documented cybersecurity capabilities, carries asymmetric risk if its safeguards can be bypassed at all, even partially. The administration's view is that the downside of being wrong is catastrophic and the cost of being cautious is temporary access disruption. Where it gets harder to defend is the lack of a statutory, transparent process for these determinations. Anthropic is not wrong that a government directive delivered at 5:21 p.m. on a Friday, without specific written justification, with no clear remediation standard, is not a sound regulatory framework for the most consequential technology of the decade.

The Broader Stakes

CNET noted that Anthropic is already in separate litigation with the Trump administration over its refusal to allow military use of its models for domestic surveillance and fully autonomous weapons systems. This shutdown lands on top of that existing legal conflict, not in isolation.

ZeroHedge flagged that Jefferies analysts see the export controls as a new headwind for Chinese AI development, since Fable 5 and Mythos 5 ranked first globally on model intelligence benchmarks and had incorporated anti-distillation features designed to limit competitors from training on their outputs.

What Comes Next

The unresolved question sitting at the center of these Washington talks is not technical. It is definitional: what remediation standard would satisfy the government enough to restore access? Anthropic's statement suggests the company has no idea what the answer is, because the directive "did not provide specific details of its national security concern." Until the administration puts a written, specific, technically grounded requirement on the table, Anthropic cannot know whether it can comply, and two of its flagship models stay dark for every customer worldwide.