28.8 Million Claude Queries: Learning Or AI Theft?
Anthropic says nearly 25,000 fake accounts generated millions of interactions in what it suspects was an attempt to distill Claude's capabilities
The AI race is no longer just about building the best models; it’s about protecting them.
According to Anthropic, nearly 25,000 fake accounts generated 28.8 million Claude interactions in just 45 days, in what the company believes may have been an attempt to distill its software engineering and agentic reasoning capabilities.
If these allegations are accurate, this represents more than platform abuse; it signals a shift toward AI intellectual property becoming a primary target.
Key Lessons
1. AI models are becoming strategic assets.
The most valuable part of an AI company may no longer be its infrastructure; it’s the intelligence encoded in its models. Protecting that intelligence is becoming as important as building it.
2. Security is now a competitive advantage.
Rate limits, identity verification, behavioral monitoring, and abuse detection are no longer optional. As AI capabilities improve, security will increasingly determine who stays ahead.
3. The AI race is expanding beyond innovation.
Competition is no longer limited to training larger models. It now includes defending proprietary capabilities, preventing unauthorized extraction, and maintaining technological leadership.
4. Distillation is a double-edged sword.
Model distillation is a legitimate machine learning technique for creating smaller, more efficient models. But when used to replicate another company’s proprietary capabilities without authorization, it raises serious legal, ethical, and competitive concerns.
5. Expect AI security to become a major industry focus.
Just as cybersecurity became essential during the internet era, AI security may become one of the defining investment areas of the next decade.
As AI systems become more capable, the challenge isn’t just creating breakthroughs—it’s protecting them.
The next phase of AI competition may be defined not only by who builds the smartest models, but by who can secure them against large-scale extraction and misuse.
What do you think?
Should AI companies prioritize open innovation, or should protecting frontier models become the industry’s top priority as capabilities continue to advance?