Anthropic disclosed this week that a China state-sponsored group used Claude to conduct large-scale automated intrusion operations, with Taiwan's defense sector among the targets, according to the company's threat intelligence report. The operation used the model to handle reconnaissance, vulnerability scanning and exploitation planning at a volume no human team matches, cutting the labor cost of sustained probing against Taiwanese networks to near zero. Anthropic said it banned the accounts and briefed government partners. The disclosure lands in the same week Beijing told Washington that Xi Jinping's September 24 White House visit depends on delaying a pending Taiwan arms sale, the move last logged on this desk September 14.

The two facts sit on the same calendar because they are the same campaign run through different instruments. One channel is diplomatic leverage tied to a summit date; the other is automated intrusion against the island's defense infrastructure, running underneath the summit talk rather than instead of it. Taiwan's Ministry of National Defense already logs PLA aircraft and ship sorties around the island daily, a public tracker Taipei has maintained through 2026. An AI-run reconnaissance layer against defense networks does not replace that pressure. It runs in parallel, on a schedule that does not pause for White House visits, and it changes what Taipei has to defend against Washington instead of only what it has to argue for.