10PB Exfiltration Claim Emerges
A claim circulated on social channels that roughly 10 petabytes were exfiltrated from a Chinese supercomputer — possibly the Tianjin National Center — but the report’s scale and veracity remain unconfirmed. The allegation, if true, would raise novel questions about the attack surface of national research infrastructure and the value of very large model training datasets. (x.com)
A person using the name FlamingChina says they pulled more than 10 petabytes from China’s National Supercomputing Center in Tianjin and has been advertising samples for sale since February 6, 2026. The claim is still unverified, which is the first thing to keep in mind before treating any of the numbers as settled fact. (news.yahoo.com) Ten petabytes is not a stack of office files. It is about 10,240 terabytes, which is large enough that even moving it slowly over months would require a sustained pipeline, not a quick smash-and-grab. (news.yahoo.com) The Tianjin center is not just one machine in one room. Its own site says it was approved in May 2009, built around the Tianhe line of supercomputers, and expanded into cloud computing, electronic government services, big data, and artificial intelligence platforms for universities, research institutes, and companies. (nscc-tj.cn) That mix matters because a supercomputing center works like a giant shared industrial workshop. One customer runs aircraft simulations, another runs climate models, another trains artificial intelligence systems, and the same campus can end up holding code, results, documents, and storage for all of them. (nscc-tj.cn) The Tianjin center’s public materials say it has petabyte-scale storage and common machine learning software such as TensorFlow and PyTorch. That means the alleged prize would not just be papers or presentations, but potentially raw datasets, model checkpoints, engineering simulations, and years of accumulated research output. (nscc-tj.cn) The best-known Tianjin system, Tianhe-1A, was installed in 2010 with 186,368 cores and ranked at 2,566 teraflops on the Linpack benchmark. In plain English, this is the kind of facility built for problems too large for normal servers, which is why its files can be unusually valuable even when they are not “classified” in the movie sense. (top500.org) The seller’s samples, according to reporting on April 9, 2026, included material tied to aerospace engineering, military research, bioinformatics, and fusion simulation. The same report said outside experts who reviewed samples for Cable News Network thought the files looked consistent with the sort of work a center like Tianjin would host. (news.yahoo.com) The reported entry point is almost as important as the alleged haul. The person behind the claim said access came through a compromised virtual private network domain, which is the kind of remote doorway organizations use so staff can reach internal systems from outside the building. (news.yahoo.com) If that detail is true, this starts to look less like breaking into one famous supercomputer and more like slipping through the side door of a sprawling research campus. Once an attacker reaches shared storage, user environments, or cloud services around the main compute cluster, the available data can be much broader than the headline machine name suggests. (nscc-tj.cn; news.yahoo.com) The reason this story spread so fast is that 10 petabytes forces a different question than most breach reports. Instead of asking whether someone stole a password database or a few secret files, it asks whether national research infrastructure has become a warehouse of training data, simulation output, and technical know-how big enough to be worth raiding at industrial scale. (nscc-tj.cn; top500.org; news.yahoo.com)
