Anthropic shut down its latest models, Fable 5 and Mythos 5, after receiving a US government notice citing “national security concerns.” What looked like a narrow export-control action quickly turned into a broader dispute over model safety, White House ties and who gains when access to advanced AI systems is cut off.
Amazon’s report helped push the issue into the White House
According to the source material, the immediate trigger came after Amazon CEO Andy Jassy alerted the White House. Amazon researchers had found a way around Fable 5’s guardrails, and a week earlier they showed officials that Claude’s protections were not flawless. Anthropic then chose to shut the models down entirely, rather than apply narrower restrictions, because the company also employs foreign nationals.
Commentators Anthony Ha, Rebecca Bellan and Sean O’Kane on TechCrunch Equity argued that the incident was about more than one compliance order. In their reading, it also exposed Anthropic’s weak relationship with the White House. O’Kane said the company’s ties with the administration were not as strong as those of other leading AI labs, describing basic communication as lacking.
Three groups were identified as beneficiaries
The first was Amazon. In the account cited by the article, Jassy’s complaint moved Anthropic’s model-safety problem from a technical finding into a political issue seen directly by the White House.
The second was cyber defenders. Several cybersecurity experts warned President Trump against keeping the restriction in place, saying the order would deprive US defenders of advanced AI detection tools. That argument gave the security community a direct stake in the outcome of the shutdown.
The third was Claude’s so-called “bad boy effect”. Rebecca Bellan said Claude downloads tend to jump whenever Anthropic is forced offline. Ramp’s data analysis was cited as support: after an earlier clash between Anthropic and the White House, Claude downloads rose noticeably. In that framing, some users who had seen ChatGPT as the default AI assistant started viewing Claude as the more resistant and more responsible alternative.
Developers in Taiwan were also affected
The impact was not confined to the US. The report said AI startups in Taiwan use Claude models as part of baseline training, and Anthropic’s shutdown left local developers unable to access the newest model APIs. That means restrictions on model exports can ripple out to teams that rely on those interfaces for ongoing development work.
The commentators also pointed to timing. O’Kane noted that the White House move on Friday afternoon came almost at the same time as US-Iran nuclear talks. Bellan said the overlap looked like an attempt to divert attention. Ha took a more restrained view, saying Anthropic’s actual risk may not be that large, but its communication problems make any risk easier to amplify.