Anthropic’s top engineers are traveling to Washington on Monday for a meeting with Commerce Department officials, a session that could determine the fate of the company’s Fable 5 and Mythos 5 models. The discussion follows a rapid escalation that began on June 9, when the firm released Fable 5 as a publicly accessible model and Mythos 5 as a restricted tool for vetted cyber defenders.

Three days later, researchers at Amazon – Anthropic’s largest investor – identified a “fix this code” jailbreak that could coax the models into producing dangerous outputs. Amazon CEO Andy Jassy relayed the findings directly to Treasury Secretary Scott Bessent, Commerce Secretary Howard Lutnick, and National Cyber Director Harry Coker Jr. That evening, Lutnick sent a letter to Anthropic CEO Dario Amodei imposing export controls on both models, citing “unacceptable risk” of diversion to foreign military intelligence users.

The shutdown and its ripple effects

By midnight on June 12, Anthropic disabled the models worldwide, a move that stunned the AI industry. The next day, Chinese AI lab Zhipu AI introduced GLM‑5.2, explicitly referencing the U.S. ban as evidence that American models are unreliable partners. Zhipu’s stock jumped 33% in a single session, underscoring the geopolitical stakes the administration feared.

More than 100 cybersecurity experts, including Stanford’s Alex Stamos and Katie Moussouris, published an open letter demanding the ban be reversed. They argue that removing the tools harms U.S. cyber defense by denying defenders the very resources they need. Meanwhile, White House officials suspect a China‑linked group accessed Mythos before the shutdown, adding another layer of concern.

Trump‑era AI adviser David Sacks claims the administration gave Amodei a clear choice: fix the jailbreak or de‑deploy the models. Anthropic disputes that characterization, and the two sides have yet to reconcile their accounts. The dispute also highlights the broader context: the Pentagon has blacklisted Anthropic as a national‑security supply‑chain threat, the company has sued the government, and the NSA continues to use Anthropic’s Claude for internal operations.

Congressional reaction spans the political spectrum. House Science Committee ranking member Rep. Zoe Lofgren said she was “appalled,” while the R Street Institute labeled the export controls a “bad idea applied badly.” A UAE‑based outlet questioned whether the Trump administration was using national‑security rhetoric to punish Anthropic.

The Monday meeting marks the first formal opportunity for de‑escalation. Sources close to the talks say communication between Anthropic and the administration has been strained, with some describing it as “speaking different languages.” A senior official called Anthropic’s handling of the known vulnerabilities “recklessness” that eroded government trust.

If the parties reach an agreement, the U.S. could avoid ceding further ground to Chinese competitors while restoring tools to its cyber defenders. Failure to resolve the dispute could prolong the suspension, threaten Anthropic’s revenue, and test the ability of Washington and Silicon Valley to solve a high‑stakes AI problem together.

Questo articolo è stato scritto con l'assistenza dell'IA.
News Factory APP - notizie agentiche per potenziare il tuo SEO e AEO.