The shutdown of Anthropic’s most advanced AI models, Fable 5 and Mythos 5, has become a proxy battle over who gets to define “AI safety”: elected officials invoking national security, or AI firms and researchers insisting the risks are being misread.

Anthropic says it disabled the models after receiving a late‑day U.S. export‑control order suspending access for “any foreign national, whether inside or outside the United States,” including its own staff, and therefore took them offline for everyone to ensure compliance.¹ The company argues the government has provided only “verbal evidence of a potential narrow, non-universal jailbreak” and that treating such a finding as grounds to recall a model “deployed to hundreds of millions of people” sets an unworkable precedent.¹ CNBC likewise notes Anthropic’s complaint that the move “does not adhere” to principles of transparent, fact‑based oversight.²

Conservative-leaning coverage emphasizes national security and China. The Washington Examiner reports that export limits were imposed “partly over suspicions that a China-linked group had accessed” Mythos 5, tying the episode to broader fears over Chinese AI advantage.³ Another piece underscores that the Commerce Department banned use by foreign nationals and that Anthropic’s suspension is framed as mere compliance with that ban, even as the firm disputes that a “narrow potential jailbreak” should halt deployment across the industry.⁴

At the same time, conservative reporting surfaces evidence that cuts against the administration’s alarm. Amazon researchers, not foreign adversaries, were “behind the jailbreak,” using prompts to find bugs in known vulnerable or synthetic code, and “no new vulnerabilities in real code were discussed,” according to security expert Katie Moussouris, who calls the response a “regulatory overreaction.”⁵

Liberal outlets foreground the power struggle between Anthropic and the Trump administration. The Guardian links the order to an escalating campaign to curb foreign access to U.S. AI and to a separate rupture after Anthropic refused to support domestic surveillance and autonomous weapons, prompting a looming Pentagon supply‑chain blacklist.¹ CNBC similarly situates the takedown in a pattern of clashes, noting the Defense Department has already labeled Anthropic a “supply chain risk” that ostensibly threatens U.S. national security.²

Across the spectrum, everyone agrees the stakes are high: export controls are no longer just about chips but about the models themselves.¹ The divide is over whether this case represents prudent early action—or a politicized, opaque test case that could chill AI innovation and security research alike.