Fable 5 Ban: The Full Behind-the-Scenes Story — Amazon Called Treasury, Not National Security

How It Actually Started — Amazon Called the Treasury Secretary

The chain of events that led to the Fable 5 ban began not with a national security agency but with a phone call from Amazon CEO Andy Jassy to Treasury Secretary Scott Bessent. Out of all of the administration officials Amazon's Jassy could have called, it was Bessent who first heard about the jailbreaking issue found in the company report. Bessent was early to sound the alarm on Mythos, worked with White House Chief of Staff Susie Wiles to re-engage the embattled company, and helped get a cybersecurity executive order across the finish line.

The choice of Bessent — a financial official rather than a defence or intelligence contact — reveals something important about how the ban unfolded. This was not a carefully considered national security operation with months of inter-agency deliberation. It was a rapid escalation driven by a single investor/partner's alarm, routed through financial channels, that triggered sweeping export controls within hours. Commerce Secretary Howard Lutnick, at the direction of President Trump, called Anthropic CEO Dario Amodei on June 12 to deliver the news.

Anthropic's Response — Engineers Fly to Washington DC

Anthropic's immediate response was operational, not just legal. The company deployed engineers to Washington DC to work directly with government officials — presenting technical evidence that the jailbreak issue had been addressed and that further safety changes were being fine-tuned. According to a US official familiar with the meetings, "the company wanted to prove everything was already resolved and further changes were being fine tuned."

But the federal Center for AI Standards and Innovation and the National Security Agency reviewed those changes and said they were not good enough. This triggered a second round of fixes — a back-and-forth between Anthropic's engineering team in DC and the government agencies reviewing the changes. The process was not fast. It ran for 20 days before all agencies were satisfied.

The Agency-by-Agency Approval Process

There was no single moment when the ban was lifted. Instead, various agency heads approved the changes gradually — one agency signing off, then another, then another — until all had cleared the changes and the models could be restored. The Axios account describes this as a piecemeal process with no single arbiter and no formal timeline. On July 1, after all agencies had signed off, the models were released globally.

OpenAI, whose own GPT-5.6 is currently in a similar restricted-preview situation, did not have visibility into the Anthropic-White House discussions and is conducting its own daily technical discussions on its model's release. The parallel is direct — but GPT-5.6 was intercepted before public launch rather than after, meaning OpenAI avoided the customer disruption Anthropic experienced while still facing the same underlying review process.

What the Story Reveals About the Current AI Regulatory System

The behind-the-scenes account makes the improvised nature of the entire process undeniable. There was no pre-existing framework for reviewing a deployed AI model's cybersecurity capabilities. There was no defined timeline for how long a review could last. There was no single agency with clear authority. The process was: one investor called one Treasury official who escalated to the Chief of Staff who triggered an export control order who suspended access for 20 days while engineers negotiated agency-by-agency in person.

This is the structural problem the voluntary standards framework being negotiated this week is designed to fix. The FT reported on July 2 that the White House is in advanced talks with OpenAI, Google, and Anthropic to finalise voluntary standards that would establish benchmarks, testing timelines, and access rules for advanced models — announced as soon as next week. The Fable 5 episode is the strongest possible argument for why that framework is needed: the alternative is what happened in June.

What remains uncertain from the Axios account: whether the Fable 5 ban actually reduced any real security risk, or whether the restrictions caused more harm than good. Cybersecurity experts wrote in an open letter to the administration that other leading AI models have the same vulnerability Amazon warned about. The ban kept Fable 5 offline for 20 days and disrupted enterprise customers worldwide for a vulnerability that already existed — unaddressed — in models that faced no restrictions at all.

Source: Axios — "How the world's top AI models were revived" July 3, 2026 · Related: July 2: Fable 5 fully restored — the complete timeline → · July 3: White House voluntary AI standards framework → · The June 12 suspension →