Alibaba Stole 29 Million AI Conversations and Nobody Noticed for Six Weeks

Alibaba used 25,000 fake accounts to steal 29 million conversations from the most advanced artificial intelligence in the world. And nobody noticed for six weeks.

This week delivered three developments that every business leader needs to understand - because each one changes the landscape your company is operating in right now.

The Largest AI Data Theft Operation in History

Anthropic sent a formal letter to the United States Senate this week accusing Alibaba of running the largest known artificial intelligence data theft operation ever recorded.

No passwords were stolen. No firewalls were broken. Alibaba used the system exactly as normal users would - at industrial scale. Twenty-five thousand fake accounts. Twenty-nine million conversations extracted over six weeks. What they were capturing was not raw data. They were extracting reasoning patterns, software engineering capabilities, and autonomous task execution logic. The intelligence of a frontier model, copied conversation by conversation.

This matters for every business using commercial AI infrastructure. The value being extracted from these systems is not the interface or the subscription. It is the accumulated reasoning capability built into the model. And that capability is now a target.

OpenAI Becomes a Hardware Company

OpenAI launched its first custom artificial intelligence chip this week. It is called Jalapeno, built in partnership with Broadcom, and it ends the complete dependency OpenAI has had on NVIDIA for inference workloads.

This is a structural shift, not a product announcement. A company that has operated entirely as a software and model provider is now building the physical infrastructure that runs its own technology. The move follows a pattern seen previously in the industry - Google built TPUs, Amazon built Trainium - and it signals that the leading AI labs view hardware control as essential to long-term cost and performance independence.

For businesses that have built strategies around the assumption that AI infrastructure costs will continue to fall, this development supports that direction. Custom silicon running purpose-built workloads is consistently more efficient than general-purpose GPU infrastructure.

The Number That Ends the Assessment Phase

For businesses across Europe still evaluating whether to move on artificial intelligence, the most important number this week comes from NVIDIA’s State of Artificial Intelligence Report.

Eighty-eight percent of enterprises say artificial intelligence has already increased their annual revenue. Eighty-seven percent say it has already reduced their annual costs. The companies still treating AI as a future consideration are now the statistical minority.

The gap between businesses that have moved and businesses that have not is no longer measured in potential. It is measured in results that are already showing up in financial performance.

What This Means for Your Business

Three stories. One direction.

AI is being targeted because its value is real and proven. The infrastructure running it is being built for permanence, not experimentation. And the majority of enterprises that have deployed it are already seeing measurable financial returns.

At Aliando, we help businesses close the gap between where they are today and where the leading companies in their sector already are.

Follow for your weekly AI intelligence. Every Tuesday.

(Do not miss the video below, where each of these developments is broken down in under a minute.)