When Washington ordered Anthropic to pull Fable 5 and Mythos 5, its most powerful AI models, from foreign users in June 2026, Chinese startup Z.ai released a near equivalent, free to download and running on Huawei chips. The crackdown became the distribution channel.
On June 13, 2026, Washington ordered Anthropic to shut its two most powerful AI models out of foreign hands. Within weeks, Chinese startup Z.ai released a near-equivalent model, free to download, running on Huawei chips. The sequence is not a coincidence. It is the mechanism the policy created.
The order, dated June 13, told Anthropic to suspend access to Fable 5 and Mythos 5 for all foreign nationals worldwide, citing national security concerns over a jailbreak vulnerability Amazon researchers had disclosed (Anthropic; BBC). Anthropic complied within days (LA Times). The shutdown was sweeping: every user outside the United States, including researchers and enterprise customers in allied countries, lost access overnight. Amazon CEO Andy Jassy had been involved in exchanges with senior US officials that preceded the order (Forbes).
Two weeks later, Beijing-based Z.ai (also known as Zhipu AI) released GLM-5-2, an open-weight model the company positioned as nearly as capable as Anthropic's Fable 5 and Mythos 5, with no access restrictions (NYT; CNBC). Independent benchmarking told the same story: GLM-5-2 topped the open-weight AI rankings while running entirely on Huawei silicon, partially bypassing the US chip export controls that had shaped the previous generation of Chinese model training (Tom's Hardware). AFR's reporting described Chinese models as closing the capability gap with Anthropic and OpenAI while undercutting them on price (AFR).
The technical category matters. An 'open-weight' model ships with its trained parameters publicly downloadable, so anyone, including an adversary, can run it locally without paying the original lab or agreeing to its terms of use (Forbes). Once a near-frontier capability is released as open weights, the only effective controls are export controls on the chips needed to run it at scale. Z.ai's choice of Huawei silicon neutralized that last lever for Chinese deployment.
The economic event matters too. When a closed frontier model is removed from a market, the displaced demand does not vanish. It migrates to the next-cheapest provider capable of meeting the use case. Z.ai's GLM-5-2 sat in exactly that slot. Forbes reported that the open-weight release put previously gated capabilities into the hands of adversaries at near-zero marginal cost (Forbes). Anthropic's earlier disabling of Fable 5 and Mythos 5 had been triggered by a US export control order related to the Amazon jailbreak research (Forbes). The intended effect was to lock down a vulnerability. The actual effect was to widen the surface area of an equivalent capability now running on non-US chips.
This is the containment paradox the policy now faces. The White House announced a broader crackdown on Chinese AI access to US technology at the end of June, signaling that the Fable/Mythos suspension was the first move, not the last (CNBC). But the response from the open-weight side keeps arriving faster than the controls. Each restriction on a closed frontier model lowers the relative attractiveness of the next open-weight release, because the displaced demand arrives pre-qualified. Z.ai did not need to out-train Anthropic. It needed to be there when Anthropic left.
The falsifier for this reading is straightforward. If independent benchmarks show GLM-5-2 is materially weaker than Fable 5 and Mythos 5 on the tasks enterprises actually care about, the story deflates into a marketing claim. The Tom's Hardware and AFR reporting suggests the gap is narrower than closed-weight labs would prefer, but the precise distance remains a measurement question rather than a settled one. Watch the next round of public benchmarks, and watch whether Anthropic restores foreign access under any condition. The answer will determine whether the Fable/Mythos shutdown is remembered as a security success or as the moment a US policy first taught the market to route around it.