← Capa | Mercado

Anthropic: Chinese Military Used Claude for Taiwan-Targeted Weapons; 151 Million Queries Fed to Alibaba

2026-11-04 · Fonte:Tom's Hardware

Anthropic: Chinese Military Used Claude for Taiwan-Targeted Weapons; 151 Million Queries Fed to Alibaba

Anthropic accuses China-linked actors of weapon development, intelligence activities and mass model distillation — including 16 air-defense suppression tools targeting Taiwan and 151 million training queries fed to Alibaba.

Anthropic has accused actors linked to China of exploiting its Claude models for weapons development, intelligence activities, and the distillation of its AI systems, according to reporting by Tom's Hardware. The allegations implicate both Chinese military researchers and major Chinese technology firms in misuse of the US frontier model, placing one of America's leading AI systems at the center of a sharpening national-security dispute.

The most consequential claims concern military applications. Researchers tied to China's armed forces allegedly used Claude to code 16 air-defense suppression tools targeting Taiwan, and to draft technical specifications for anti-torpedo systems. The reported activity would mark a direct application of a US-built frontier model to weapons-adjacent engineering work aimed at Taiwan, one of the most sensitive flashpoints in US-China relations.

The alleged abuse extends beyond weaponry into model extraction. Anthropic says 151 million training queries were fed to Alibaba, a volume consistent with systematic distillation — the practice of harvesting one model's outputs to train or improve another. Chinese technology giants were also caught using Claude, per the report, indicating the misuse reached well beyond state-linked military users into commercial AI development.

Taken together, the disclosures describe a frontier model being pressed into service for weapons engineering, intelligence work, and large-scale knowledge transfer to a Chinese rival. For Anthropic, the findings test the limits of the usage policies and detection systems that govern who can access frontier capabilities and how abuse is identified after the fact.

For AI-compute buyers and investors, the case underscores that frontier-model security — access controls, usage monitoring, and anti-distillation defenses — is becoming a differentiator for US labs, and that acceptable-use enforcement now carries direct geopolitical and commercial weight across the AI supply chain.

Tags: Anthropic · Claude · Alibaba · Model Distillation · AI Safety · China


Leituras relacionadas


萬安算交所 AIXX · AIXX · Exchange verificado · Ir ao mercado