China Steals, Then Copies: How Beijing Is Racing to Match America's Most Dangerous AI

.

A Weapon Too Powerful to Ignore

When Anthropic unveiled its Mythos AI model in April, the reaction in Washington was immediate alarm — and in Beijing, something close to panic.

Mythos is not an ordinary AI assistant. It can automatically scan software for security vulnerabilities — the hidden flaws that hackers exploit to break into systems. Cybersecurity experts warned almost immediately that such a tool could supercharge cyberattacks on a scale never seen before. The U.S. government moved fast, ordering Anthropic to suspend exports of even a less powerful version of the program, citing national security concerns.

China was watching closely.

360's Answer: "China Cannot Afford to Wait"

On Wednesday, at the ISC.AI 2026 cybersecurity conference in Beijing, Zhou Hongyi — founder of Chinese cybersecurity giant 360 Security Technology and a member of China's top political advisory body — took the stage with a pointed message.

He unveiled two new AI security tools under the banner "Yitian Tulong," a name borrowed from a classic Chinese martial arts novel meaning "Heavenly Sword and Dragon Saber."

The first tool, called "Tulongfeng," is designed to automatically discover software vulnerabilities. Zhou called it directly "China's version of Mythos." The second tool, "Yitianzhen," focuses on automated cyber defense and incident response.

"This kind of powerful weapon that can change the landscape of cyber offence and defence cannot be held only by others," Zhou told the audience, according to a transcript published by his company.

Zhou framed the issue as a matter of national survival. He argued that China faced a dangerous condition of "one-way transparency" — a situation where U.S. entities could use Mythos-like tools to scan Chinese software and critical infrastructure, while Chinese companies had no comparable capability to respond in kind.

360 claims Tulongfeng has already found 3,432 software vulnerabilities, including 105 confirmed by Chinese authorities. Reuters said it could not independently verify these figures.

Zhou also acknowledged that Chinese AI models still lag behind American ones by around 20 to 30 percent in raw capability — a gap he attributed in part to U.S. export controls on advanced chips in place since 2022. His solution: don't wait for the models to catch up. Instead, 360 is combining AI with specialized security expertise, vulnerability databases, and automated tools — what he calls an "agent" approach.

"If Mythos is a top-end chip, what we are building is a complete machine that can run stably, work 24 hours a day and make fewer mistakes," he said.

The Theft Behind the Announcement

While Zhou was presenting China's homegrown answer to Mythos, a very different story was unfolding behind closed doors in Washington.

On June 10 — two weeks before Zhou's announcement — Anthropic had sent a letter to two senior U.S. senators detailing what it described as the largest known attack of its kind on the company. The target: Claude, Anthropic's flagship AI platform. The attacker: operators affiliated with Alibaba, China's e-commerce and technology giant.

Reuters obtained and reported on the letter on Wednesday evening.

According to Anthropic, the operation ran from April 22 to June 5, 2026 — a period of roughly six weeks. During that time, nearly 25,000 fraudulent accounts were used to generate more than 28.8 million exchanges with Claude. The goal, Anthropic said, was "distillation" — a technique where a less capable AI model is trained on the outputs of a more powerful one, essentially copying its intelligence without building it from scratch.

Anthropic stated clearly that this distillation campaign was designed to help accelerate China's ability to reach the advanced capabilities of Mythos Preview.

Alibaba did not respond to a request for comment. The company was added to the Pentagon's list of Chinese military companies earlier this month — a designation Alibaba is currently challenging in court.

A Pattern of Escalation

The Alibaba case is not isolated. Anthropic revealed in February 2026 that it had already identified similar campaigns by three other Chinese AI companies: DeepSeek, Moonshot AI, and MiniMax.

DeepSeek — whose budget AI model rattled Silicon Valley in early 2025 — was involved in over 150,000 exchanges. Moonshot AI conducted over 3.4 million. MiniMax topped 13 million. The Alibaba operation, at more than 28.8 million exchanges, dwarfs all of them.

Anthropic warned in February that these campaigns were growing in "intensity and sophistication" — a warning that now looks prescient. The company said at the time that stopping the threat would require rapid, coordinated action from industry, policymakers, and the international AI community.

In April, the White House publicly accused China of stealing U.S. AI intellectual property on what it called an industrial scale.

Washington's Response — and Its Contradictions

The U.S. government's response has been swift in some areas and hesitant in others.

On June 12 — just two days after Anthropic sent its letter about Alibaba — the Commerce Department imposed sweeping restrictions on Anthropic's Mythos and Fable AI models, citing fears they could be accessed by military intelligence users in China and other countries of concern. Anthropic was forced to disable global access to the models entirely.

Yet in a separate move that raised eyebrows, the Commerce Department has so far held off on placing DeepSeek on its trade blacklist — despite an interagency committee reportedly finding it poses a national security risk. Officials appear to be trying to avoid further escalating tensions with Beijing.

The result is a policy landscape that is, at best, inconsistent: restricting American AI exports to allies and neutral countries while hesitating to penalize the Chinese companies accused of stealing that same technology.

What It All Means

Taken together, the two stories from June 24 paint a clear picture of where the global AI race stands.

China is working on two tracks simultaneously: building domestic alternatives to American AI tools — and stealing the capabilities it cannot yet replicate. The 360 announcement and the Alibaba distillation campaign are not contradictory. They are complementary strategies in a single effort to close the gap with the United States as fast as possible.

Anthropic said it supports government efforts to combat these attacks, including through threat-intelligence sharing with the private sector. But with campaigns of this scale now confirmed, the question is whether reactive measures are enough — or whether the window to act is already closing.

.

Sources:

1. Reuters – China's 360 says it has developed tools to match Anthropic's Mythos (June 24, 2026): https://www.reuters.com/legal/litigation/chinas-360-says-it-has-developed-tools-match-anthropics-mythos-2026-06-24/
2. Reuters – Anthropic says Alibaba illicitly extracted Claude AI model capabilities (June 24, 2026): https://www.reuters.com/world/china/anthropic-says-alibaba-illicitly-extracted-claude-ai-model-capabilities-2026-06-24/
3. Reuters – Chinese companies used Claude to improve own models, Anthropic says (February 2026): https://www.reuters.com/world/china/chinese-companies-used-claude-improve-own-models-anthropic-says-2026-02-23/
4. Reuters – White House accuses China of industrial-scale theft of AI technology (April 2026): https://www.reuters.com/world/white-house-accuses-china-industrial-scale-theft-ai-technology-ft-reports-2026-04-23/
5. Reuters – U.S. blocks foreign access to Anthropic's most advanced AI models (June 13, 2026): https://www.reuters.com/technology/us-blocks-foreign-access-anthropics-most-advanced-ai-models-axios-reports-2026-06-13/

.