Anthropic has accused Alibaba of using nearly 25,000 fraudulent accounts to extract capabilities from its Claude artificial intelligence models, an alleged campaign the US company described as the largest known attack of its kind against it.
According to a June 10 letter Anthropic sent to senior members of the US Senate Banking Committee, the operation ran from April 22 to June 5 and generated more than 28.8 million exchanges with Claude. The company said the activity involved distillation, a technique in which a less capable model is trained on the outputs of a more advanced system, potentially allowing a rival to replicate some capabilities at lower cost.
Anthropic said the campaign was conducted by operators affiliated with Alibaba and Alibaba Qwen, the Chinese technology giant's AI lab. Alibaba did not immediately respond to a request for comment. The allegations have not been independently verified, and Alibaba has not confirmed the details in the letter.
Why distillation matters
Distillation is not new. It has long been used to compress large models into smaller ones that run more cheaply or on less powerful hardware. In a legitimate setting, a developer trains a student model on outputs from a teacher model, aiming to preserve useful behavior while reducing cost. When that process is carried out without authorization through a commercial API, it becomes a form of model extraction.
The commercial logic is straightforward. Building and training a frontier AI model can require immense computing power, large datasets, and specialized engineering. Accessing a model through an API costs far less. If a competitor can collect enough high-quality outputs, it may approximate parts of the original model's behavior without bearing the full cost of development. The result is rarely an exact copy, but even partial replication can erode a model maker's advantage.
Fake accounts make extraction harder to detect. Attackers can spread requests across thousands of identities, rotate prompts, and vary usage patterns to avoid rate limits. A campaign of 28.8 million exchanges over roughly six weeks implies a high degree of automation and persistence. Vendors must distinguish that behavior from legitimate enterprise workloads, which can also be large and complex.
US-China technology tensions
The dispute lands in the middle of growing US-China technology competition. AI development is increasingly tied to export controls, semiconductor restrictions, and national security reviews. Anthropic said the alleged campaign could help accelerate China's ability to reach the capabilities of its advanced Mythos Preview model. US officials have stepped up scrutiny of advanced AI systems over fears they could be used by military or intelligence users in countries of concern.
That context raises the stakes. Model extraction is not only a commercial intellectual property issue. It also touches safety and security. If a model's behavior can be copied outside its original safety controls, guardrails may not transfer. A cloned or distilled system could be deployed without the same monitoring, usage policies, or security architecture that surrounded the original.
Regulators may also see the allegations as evidence that AI supply chains need more oversight. The letter went to the Senate Banking Committee, a panel whose interest in AI has grown as financial institutions adopt the technology. Banks and critical infrastructure operators may be asked to assess AI suppliers more like other critical vendors, with stronger due diligence and contingency planning.
Earlier campaigns
Anthropic said in February that it had identified similar campaigns by DeepSeek, Moonshot AI, and MiniMax to extract capabilities from Claude. The alleged activity ranged from more than 150,000 exchanges by DeepSeek to more than 13 million by MiniMax. The latest alleged campaign, at 28.8 million exchanges, is larger. Analysts see that pattern as escalation rather than an isolated risk.
These disclosures matter because API access is central to the business model of many AI vendors. Customers pay for usage, and vendors set limits to prevent abuse. When attackers use many fake accounts, detection becomes harder. Vendors must balance friction against legitimate enterprise use. Too little friction invites abuse; too much can drive customers away.
Enterprise supply-chain risk
If Anthropic's claims are true, the alleged campaign could allow Alibaba to build a comparable model in a short period of time and offer it at a much lower cost, said Anand Joshi, an AI analyst at TechInsights. Joshi said enterprise buyers should ask what training data was used, how the model was trained, what guardrails exist, and how they can audit it. He also said model publishers will need watermarking technology in models and model responses so that if skills are stolen, they can find the thief.
Sanchit Vir Gogia, chief analyst at Greyhound Research, said the enterprise supply chain no longer ends at software, APIs, and cloud regions. It now includes rented intelligence, and rented intelligence can be copied and redeployed well outside the safety controls it was born with. Gogia said distillation should be a board-level concern because a weaker model trained on a stronger one can inherit capabilities without the governance and controls around the original system.
For enterprises, the allegations point to a potentially more serious risk than conventional intellectual property theft: reverse engineering at scale. If proven, they would suggest that AI models can be copied systematically, turning model extraction into a new AI supply-chain risk. Pareekh Jain, CEO of Pareekh Consulting, said if a rival can clone the exact brain of the AI a company relies on, they can easily find its blind spots, hack automated systems, or cause the AI vendor to panic and shut down services that a business needs every day.
Mitigating the risks
The allegation raises questions about the controls AI vendors have in place and how customers can protect themselves. Jain said vendors should provide verified accounts, smart rate limits, abuse detection, usage monitoring, contractual bans on distillation, incident disclosure, and audit rights. Enterprises should ask how a vendor detects and blocks large-scale model extraction and can demand contracts that guarantee backup plans and financial refunds if the AI service is attacked or suddenly shut down.
Those demands may become standard in procurement. Companies increasingly embed AI in customer service, software development, security operations, and financial workflows. An outage or degraded service is not just an inconvenience. It can halt operations. If a model is compromised or withdrawn due to abuse, customers need continuity plans and clear remedies.
Vendors also face regulatory pressure. Lawmakers are examining AI risks, including national security, data protection, and competition. The letter to the Senate Banking Committee signals that model extraction is being framed as a financial and security issue, not only a technical one. That framing could lead to new expectations for AI providers, especially those serving regulated industries.
Detection challenges
Large-scale extraction is hard to detect because it mimics legitimate usage. Attackers can rotate accounts, vary prompts, and distribute traffic across regions. Defenses include identity verification, anomaly detection, prompt pattern analysis, and limits on bulk outputs. But aggressive controls can hurt legitimate developers. Vendors must calibrate thresholds and investigate suspicious behavior without blocking enterprise customers.
Watermarking and fingerprinting are promising but immature. A model may be trained on another model's outputs, making provenance difficult to prove. Contracts and audit rights can help, but enforcement across borders is complex. The alleged campaign also highlights the role of cloud and API intermediaries. If accounts are created with false identities, vendors may need stronger know-your-customer checks and better coordination across services.
What happens next
Alibaba has not responded publicly to the specific claims. Anthropic's letter is not a legal finding. The Senate Banking Committee may seek more information, and the allegations could prompt broader scrutiny of how AI labs prevent model extraction. The dispute also could intensify calls for international rules on AI model copying and security.
For now, the key facts are these: Anthropic alleges nearly 25,000 fake accounts, more than 28.8 million exchanges, a campaign from April 22 to June 5, and involvement by operators affiliated with Alibaba and Alibaba Qwen. The company calls it the largest known attack of its kind. The allegations remain unresolved, but they have already pushed model extraction into boardroom and policy discussions.
Source: InfoWorld News