Nvidia CEO Jensen Huang has warned that China already has the computing power and data center performance necessary to train an AI model at the same level as Claude Mythos’ Anthropic AI model, which could threaten global cybersecurity.
Huang was asked in interview on Wednesday on Dwarkesh Patel’s podcast, whether the Chinese government’s access to chips for training a model like Claude Mythos – who has cyber-offensive capabilities – could pose a threat to US national security.
Mythos was trained in “pretty mundane skills,” Huang said.
“The amount of capacity and the type of calculations he was trained on are widely available in China, so you first have to realize that the chips exist in China.”
In April, Anthropic restricted access to its novel artificial intelligence model after identifying thousands of software vulnerabilities in major operating systems and browsers, raising concerns about potential abuse in cyberattacks. A Chinese-made AI model with the same capabilities could wreak havoc if misused.
Huang said the amount of computing China has is “enormous.”
“They have data centers that are completely empty, fully powered. You know, they have ghost cities, they also have ghost data centers. They have so much infrastructure capacity. If they wanted to, they would do it [could] just collect more chips.”
A call for dialogue, not conflict
Huang added that China produces 60% of the world’s major chips, employs some of the best computer scientists, 50% of the world’s artificial intelligence researchers and a lot of energy.
“Victimizing them and turning them into the enemy is probably not the best answer,” he said. “They are the enemy.”
“We want the United States to win. But I think having a dialogue and a research dialogue is probably the safest thing to do.”
Related: Anthropic restricts access to the AI model due to concerns about cyberattacks
On Tuesday, US Treasury Secretary Scott Bessent hailed Mythos as a revolutionary step that will put America ahead of China in the artificial intelligence race. “This Anthropic Mythos model was a step function of change in ability and learning,” he said. According to to Bloomberg.
Claude Mythos poses a real threat
Anthropic published findings on the Claude Mythos Preview on April 7, raising concerns that the model could be used in cyberattacks due to its ability to discover and potentially exploit zero-day vulnerabilities. The company also claimed that 99% of the vulnerabilities discovered by the model have not yet been patched.
Meanwhile, the AI Security Institute (AISI) rated Mythos on April 13, stating that the AI model can “perform multi-stage attacks on vulnerable networks and discover and exploit vulnerabilities on its own,” which would take professionals days of work.
Artificial intelligence hacks using Mythos could also have dire consequences for banks, which often employ decades-old software, Reuters reported on Tuesday.
Last year, Anthropic reported in November that a “Chinese state-sponsored group” had manipulated the Claude Code tool in an attempt to infiltrate approximately 30 targets around the world, with success in a compact number of cases.
Warehouse: How Artificial Intelligence Dramatically Accelerated Quantum Risk for Bitcoin
