The White House classified Anthropic's most capable models as too sensitive for continued commercial deployment this week. Reporting placed PRC actors inside the Mythos AI system before that order was signed. Amazon CEO Andy Jassy had already briefed the administration on Anthropic's risk profile. That detail matters. Jassy runs the $4 billion AWS compute relationship that funds Anthropic's training runs, which positioned him as the investor recommending restriction of his own portfolio company's primary revenue asset.

OpenAI filed a confidential S-1 with the SEC this week. Anthropic cannot. The split between the two US labs is now regulatory, not architectural: both trained on NVIDIA B200 NVL72 hardware, on silicon that BIS October 2023 export controls deny to PRC buyers. If the Mythos breach report holds, the PRC was running inference or fine-tuning on extracted Anthropic architecture through Ascend 910B clusters in Suzhou and Beijing hyperscale facilities at the moment the White House was preparing to classify that architecture as off-limits for domestic commercial deployment. BIS October 2023 governs forward export, not weights already outside the perimeter; the restriction, effective the week of June 15, 2026, arrived after the window it was written to close.