Basiran
In a stunning escalation of the escalating geopolitical "AI arms race," artificial intelligence firm Anthropic has leveled serious allegations against Chinese tech giant Alibaba. According to a confidential report shared with U.S. lawmakers, Anthropic claims that operators affiliated with Alibaba and its specialized AI lab, Alibaba Qwen, executed a sophisticated, industrial-scale campaign to illicitly extract the core capabilities of its frontier model, Claude.
The incident, which Anthropic describes as the "largest campaign to illicitly extract Claude’s capabilities ever measured," highlights the growing desperation among international players to bridge the gap between their own domestic models and the advanced, high-performance systems currently dominating the U.S. market.
The Scope of the Alleged Infiltration
The data provided by Anthropic paints a picture of a calculated, persistent operation. Between April 22 and June 5, 2026, the company recorded more than 28.8 million individual exchanges with its Claude model. These interactions were not the result of legitimate user traffic; rather, they were funneled through nearly 25,000 fraudulent accounts designed to bypass Anthropic’s standard security protocols.
The objective of this massive data-harvesting effort was not merely to converse with the model, but to perform what industry experts call "distillation." By repeatedly prompting the model and observing its responses, the attackers sought to reverse-engineer its internal logic. Anthropic reports that the campaign specifically targeted Claude’s most sophisticated functions, including agentic reasoning—the ability of an AI to perform complex, multi-step actions autonomously—advanced software engineering, and long-horizon planning.
Anthropic engineers noted that the attackers employed sophisticated obfuscation techniques, including the use of complex proxy networks and rotating IP addresses, to mask the origin of the requests. This suggests a level of technical coordination consistent with state-backed or large-enterprise-level operations.
A Chronology of Escalation
The timeline of these events suggests a direct defiance of tightening U.S. trade and security policies.
- April 2026: The U.S. government, under the Trump administration, formally accuses several Chinese labs—including DeepSeek, Moonshot, and MiniMax—of "industrial-scale" AI theft. Anthropic reveals that these firms had previously attempted to extract data from Claude using approximately 24,000 accounts.
- April 22, 2026: The alleged Alibaba-linked campaign commences, marking a continuation of the same tactics despite the high-profile warnings from Washington.
- June 5, 2026: The campaign concludes, having racked up nearly 29 million exchanges.
- June 10, 2026: Anthropic sends a confidential letter to Senators Tim Scott (R-SC) and Elizabeth Warren (D-Mass.), providing evidence of the breach just one day before a high-stakes Senate committee hearing regarding the future of the "American Dream" in the age of AI.
- Mid-June 2026: Public discourse intensifies as tech leaders in China, such as 360 Security Technology founder Zhou Hongyi, openly advocate for the development of "cyber nuclear weapons" to match U.S. advancements.
The Growing "Circumvention Economy"
Anthropic’s report highlights the emergence of what it terms a "circumvention economy." As the U.S. implements stricter export controls on high-end GPUs—such as those produced by NVIDIA—Chinese firms are finding it increasingly difficult to train frontier-grade models from scratch. Consequently, they are turning to distillation: the practice of using existing, superior models to train their own smaller, domestic models.
This "distillation attack" model is becoming a significant concern for U.S. national security. By essentially "stealing" the intelligence of U.S. models, foreign firms can leapfrog years of costly research and development. Anthropic warns that this effectively transforms hundreds of billions of dollars of American private and public R&D investment into a massive subsidy for geopolitical competitors.
Official Responses and Diplomatic Friction
Alibaba has vehemently denied any illicit activity. In a legal filing made public recently, the company challenged its designation as a "Chinese military company" by the U.S. Department of Defense, calling the claim baseless.
"Alibaba is governed by an independent board, none of whom has any military affiliation," the company stated in a press release. "Our products and services are built for retail, logistics, and enterprise information technology—not weapons, defense, or intelligence."
Despite these denials, the political climate remains hostile. Anthropic, in its correspondence with Congress, urged for a multi-pronged legislative response:
- Antitrust Modernization: Allowing AI firms to share threat intelligence about foreign actors without fear of running afoul of antitrust regulations.
- Expanded Export Controls: Tightening the supply chain for advanced compute to ensure that firms cannot acquire the hardware necessary to host or train on stolen frontier model weights.
- Strict Penalties: Implementing legislative consequences for companies found to be utilizing stolen model capabilities, including banning them from the U.S. market or restricting their access to U.S.-based data centers.
Implications for Global AI Governance
The broader implication of this incident is the potential for a "decoupled" AI ecosystem. If the U.S. continues to restrict access to its frontier models, and Chinese firms continue to attempt to circumvent these restrictions through clandestine data extraction, the world may soon face two distinct, incompatible AI standards.
Anthropic warns that if China succeeds in cloning these capabilities, they could potentially deploy them for cyber-offensive operations against the U.S. government and private sector. "The larger the capability gap between our models and theirs, the more time the U.S. government has to harden our cyber defenses," an Anthropic spokesperson noted. "If that gap closes, we face a future where adversaries can exploit our infrastructure faster than we can patch it."
Furthermore, the lack of safety guardrails in "bootlegged" versions of these models is a major point of contention. Anthropic fears that cloned versions of their models, stripped of their internal safety mechanisms, could be used for malicious purposes, such as generating biological threats or facilitating widespread automated disinformation campaigns.
A "Cyber Nuclear Weapon"
The severity of the situation was underscored by the recent comments of Zhou Hongyi, the founder of 360 Security Technology. Speaking at a conference in Beijing, Zhou characterized the U.S. lead in AI—specifically the Mythos model, which demonstrated an unprecedented ability to identify zero-day vulnerabilities—as a "cyber nuclear weapon."
Zhou’s admission that China is currently "well short" of these capabilities is telling. He argued that for China, the only path to national security is to develop a domestic equivalent. "If we do not have our own, we are at the mercy of those who do," he warned.
This rhetoric has fueled a sense of urgency within the Chinese tech industry, leading to massive investments in local models like the Qwen series. However, the reliance on U.S. infrastructure and the temptation to take "shortcuts" via distillation continue to define the current conflict.
Conclusion: The Long Road Ahead
As the dust settles from the latest accusations, one thing is clear: the era of "open" AI development is coming to an abrupt end. The race for technological supremacy has transformed into a high-stakes intelligence game where every interaction with a chatbot is now a potential battlefield.
Whether Congress will act on Anthropic’s recommendations remains to be seen. However, as the geopolitical tension between the U.S. and China continues to bleed into the realm of algorithms and neural networks, the cost of innovation is no longer just measured in dollars—it is measured in security, stability, and the ability to maintain a technological edge in an increasingly digital world.
For Alibaba, the immediate future will be defined by its legal battle with the U.S. government and its attempt to maintain its status as a global tech leader despite the mounting accusations of espionage. For the rest of the world, the incident serves as a stark reminder that in the age of AI, the most valuable assets are not just the data themselves, but the systems that know how to use them.
Siti Muinah
Dwi Wanna
