Anthropic Accuses DeepSeek of AI Model Theft

Anthropic just accused three Chinese AI labs of stealing its technology

Anthropic, the company behind Claude, revealed that DeepSeek, Moonshot AI, and MiniMax ran coordinated campaigns to extract Claude’s capabilities using roughly 24,000 fraudulent accounts and 16 million exchanges. The technique is called model distillation — training a cheaper model on the outputs of a more capable one.

This is not a theoretical risk. It happened at industrial scale, and it has real implications for the AI tools your business depends on.

What happened

The attacks

Each lab targeted different Claude capabilities:

• DeepSeek ran over 150,000 exchanges focused on reasoning tasks, reward model functions, and generating alternatives to politically sensitive queries.
• Moonshot AI drove over 3.4 million exchanges targeting agentic reasoning, tool use, coding, and computer vision.
• MiniMax was the most aggressive — over 13 million exchanges concentrating on coding and tool orchestration capabilities.

In one case, a single proxy network managed more than 20,000 accounts simultaneously, mixing distillation traffic with normal customer requests to avoid detection.

How Anthropic caught them

Anthropic traced the campaigns through IP address correlation, request metadata, and infrastructure indicators. In some cases, they tracked accounts back to specific researchers at the labs. Industry partners confirmed the same actors were targeting their platforms too.

OpenAI separately reported similar activity from DeepSeek in an open letter to U.S. legislators.

What model distillation is and why it matters

Distillation is a legitimate AI training technique. Frontier labs use it internally all the time — it is how companies create smaller, faster models from their most capable ones. The problem starts when competitors use it to copy capabilities they did not build.

Think of it like this: if a competitor sent 24,000 fake customers into your restaurant to reverse-engineer every recipe, that would be industrial espionage. The AI version works the same way, just at machine speed.

The concern goes beyond intellectual property. Models built through illicit distillation typically lack the safety guardrails of the originals. As Anthropic put it: “These campaigns are growing in intensity and sophistication. The window to act is narrow.”

How AI IP disputes could affect small business tool pricing

If you use AI tools in your business — for customer service, scheduling, content creation, inventory management — this fight matters to you. Here is why.

Higher costs to fund defense

Building systems to detect and prevent distillation attacks costs money. Anthropic now runs behavioral fingerprinting systems, enhanced verification, and output safeguards specifically to counter these threats. Those costs get built into pricing. Every AI provider dealing with this — and OpenAI and Google face similar attacks — passes some of that cost downstream.

Tighter API restrictions

Expect stricter terms of service, usage limits, and account verification across AI platforms. If you use AI APIs directly or through tools like Appalach.AI’s AI Employees, you may see more authentication steps and usage monitoring. This is a necessary trade-off for tool reliability.

Cheaper knockoffs with hidden risks

Distilled models strip away safety testing. A budget AI chatbot built on stolen capabilities might seem like a bargain, but it could generate harmful content, leak customer data, or produce unreliable outputs. When evaluating AI tools for your business, knowing where the model comes from matters.

Market consolidation

Smaller AI companies that cannot afford distillation defenses may get squeezed out. That could mean fewer choices and less competition — never great for small business buyers. The companies that survive this phase will be the ones with strong infrastructure and defensible technology.

What to watch as the AI competition escalates

Policy is catching up

This disclosure came as the Trump administration allowed U.S. companies to export AI chips to China, a decision critics argue undermines the very protections Anthropic is calling for. Expect more regulatory action — possibly new export controls, AI IP protections, or mandatory security standards for AI providers.

If you are curious about the broader AI safety landscape, we covered how global frameworks affect small business tools in our post on the UK’s AI safety push.

What you should do now

1. Know your AI supply chain. Ask your AI tool vendors where their models come from and what safety testing they undergo. Legitimate providers will answer clearly.
2. Be skeptical of too-cheap AI tools. If an AI product offers frontier-level capabilities at rock-bottom prices, ask how. The answer might be distillation from a provider that invested billions in research.
3. Diversify your tools. Do not build critical workflows on a single AI provider. If a distillation scandal or regulatory action disrupts one platform, you want alternatives.
4. Stay informed. AI policy is moving fast. Decisions made in Washington, Brussels, and Beijing this year will shape the tools available to your business next year.

The bottom line

The Anthropic-DeepSeek conflict is not just a tech industry drama. It is a signal that AI development is entering a phase where intellectual property, national security, and market dynamics are colliding. For small businesses, the practical impact comes down to pricing, tool reliability, and the importance of choosing AI partners who build responsibly.

The companies that invest in original research and robust safety will produce more reliable tools. That is the bet we make at Appalach.AI — building on trusted AI infrastructure so your business gets tools that work as advertised. If you are evaluating AI tools for your business, explore our services or get in touch to talk through your options.