Skip to main content

icnlive

WATCH LIVE. THINK BUSINESS.

© 2026 ICN.LIVE

PUBLISHED BY ICN PRESS

  • Anthropic accused Alibaba of the largest known distillation attack against its Claude AI models.
  • Nearly 25,000 fake accounts generated 28.8 million Claude exchanges between April and June 2026.
  • The campaign links to Alibaba and its Qwen AI research unit, Anthropic says.
  • US regulators later restricted Anthropic’s newest Mythos and Fable models over security fears.

AI theft accusations from Anthropic now target Alibaba over a sweeping campaign against Claude AI models. The American firm told US senators about the breach in a June 10 letter. Anthropic claims the secret operation ran between April 22 and June 5 this year. Nearly 25,000 fraudulent accounts produced more than 28.8 million exchanges with the Claude system. Those numbers make this the largest known Claude AI distillation attack on the company. Senators received the warning letter ahead of a planned hearing on artificial intelligence safety. Anthropic shared the detailed figures to clearly show the unusual size of this breach.

Inside the Alibaba Qwen Claude extraction claim

Distillation trains a weaker model on the outputs of a stronger and smarter one. Engineers feed millions of answers into a cheaper system to copy advanced skills fast. Anthropic says the operation aimed to speed China’s race toward its Mythos Preview models. The company tied the whole effort to Alibaba and its Qwen AI research unit. Operators behind the Alibaba Qwen Claude extraction targeted software engineering and agentic reasoning skills. Both of these areas rank among the most valuable frontiers in modern AI today. Anthropic argues this method lets rivals copy advanced systems without the original research cost.

What the AI theft accusations from Anthropic involve

Earlier this year, Anthropic also flagged similar attempts by three other Chinese AI labs. DeepSeek generated over 150,000 exchanges, while Moonshot AI passed 3.4 million with the model. MiniMax then topped 13 million exchanges, making Alibaba’s alleged campaign much larger by comparison. American firms now warn loudly about AI intellectual property theft tied to China today. From my standpoint, this scale of extraction signals a serious shift in competitive pressure. These campaigns grow more advanced each year, the company warned policymakers and cloud partners.

US AI export controls now limit how China reaches Anthropic’s newest and strongest systems. Commerce officials restricted the latest Mythos and Fable models on June 12 this year. Anthropic then disabled global access to those models after the new government directive arrived. In its letter, Anthropic wrote that Alibaba “ignored the Trump Administration’s warnings” during the campaign. Officials also added Alibaba to the Pentagon’s list of Chinese military companies this month. Alibaba is challenging the designation while expanding its Qwen AI work across global markets. Washington has openly accused Beijing of stealing US AI ideas on an industrial scale. Regulators held off blacklisting DeepSeek even after flagging the firm as a security risk.

Why the Anthropic-Alibaba AI theft fight matters to you

These AI theft accusations from Anthropic shape how nations guard their best AI research. Models built through copied outputs often skip the safety checks of their original source. For everyday users, weaker safety controls raise real risks of misuse and false output. Independent experts say these copied models can spread harmful content with fewer working safeguards. Anthropic wants closer threat sharing between AI developers and the US government going forward. Lawmakers will weigh these AI theft accusations from Anthropic during upcoming hearings on policy. Your trust in safe AI tools now depends on how this serious case ends. For now, the AI theft accusations from Anthropic keep global AI competition firmly tense.

TAGS

EXPLORE MORE ON:

6G band in the UAE

The case for the 6G band in the UAE rests on a small number: two to five percent. That is how far an Upper 6 GHz signal falls behind today’s 3.5 GHz C-band in reach. The modelling covered city centres, towns, suburbs and open country. Higher frequencies normally travel less far, and shorter reach normally means more towers. A gap this narrow points the other way. An operator can hang the new radios on masts and rooftops it already owns.

The figure comes from a white paper titled “The Golden 6 GHz Band.” Four names sit on it: TDRA, the telecoms regulator, e& UAE, Khalifa University of Science and Technology, and NYU Abu Dhabi. Its release comes with a decision to move the 6G band in the UAE out of trials and into commercial service. The industry calls it the golden spectrum.

Where the threefold figure comes from

Two gains sit behind the headline claim, and they multiply. The first is a gain in spectral efficiency of about 1.5 times. It comes from a 256TRX Giga-MIMO antenna array, which has far more transmit and receive paths than a standard 5G unit. Beamforming and scheduling across the array do the rest. This gain holds regardless of the channel width. The second is width itself. Set 200 MHz of Upper 6 GHz spectrum against a 100 MHz C-band carrier and the channel doubles. One and a half times two comes to roughly triple the capacity per cell. The authors say the result lines up with work by independent analysts.

Coverage comes out of the same hardware. Signals at higher frequencies lose strength faster, and the array’s beamforming makes up for the loss. In the hardest case the team modelled, indoors in a dense city, reach stays within about three percent of C-band.

A caveat travels with every figure. The results come from models, not field tests, and rest on assumptions the paper lists. Live performance, it says, will depend on where the radios go, which spectrum they get and what the devices can do. Bayan Sharif, provost of Khalifa University, called the 1.5 times gain “achievable under well-conditioned assumptions.” The method went out with the findings, he said, so others could check the work.

Business Intelligence & News

  • Meta One in the UAE brings paid tiers from Dh5.99 to Dh1,199 a month across WhatsApp, Instagram, Facebook, and Meta AI.
  • Core versions of all four apps stay free, with subscriptions adding features and heavier AI usage on top.
  • Entry plans start at Dh5.99 for WhatsApp Plus, while creator and business bundles climb into the hundreds.
  • Meta reports more than 15 million subscriptions and trials worldwide, with Edits and AI glasses features still to come.

How regulation shaped the 6G band in the UAE

Policy moved before engineering on the 6G band in the UAE. In 2023, the World Radiocommunication Conference identified 6425 to 7125 MHz for mobile service across ITU Region 1. The Telecommunications and Digital Government Regulatory Authority (TDRA) moved earlier than most of its peers and wrote the whole range into its national frequency plan. Within it, e& UAE holds 6425 to 6775 MHz, as much as 350 MHz in one unbroken block.

With the allocation settled, the operator committed to a commercial Giga-MIMO deployment built for peak downloads of 10 Gbps. Launch is planned between July and December 2026. Tariq Al Awadhi, TDRA’s executive director of spectrum affairs, drew the line himself: “Regulatory certainty is what turns research into infrastructure.”

Money follows the coverage result. The white paper treats Upper 6 GHz as a capacity layer laid over e& UAE’s current 5.5G network. It shares towers, rooftops, power and transport links, and backhaul gets an upgrade where needed. Fewer new sites can mean a lower cost for each bit carried as demand grows. Marwan Bin Shakar, chief technology officer at e& UAE, framed the customer side as higher speeds, more capacity and a steadier connection in crowded areas.

What the network is meant to carry

The paper groups planned uses under three headings: Connect Home, Connect Industry, and Connect Consumer and Vehicle. Those cover home broadband delivered over the air at speeds close to fibre, heavy-bandwidth uses for companies and public bodies, AI-driven services in the home, and connected cars.

The work extends earlier UAE research, including TDRA’s national 6G roadmap and two e& UAE papers written with the same universities. TDRA has described the band as a resource for 5G-Advanced services and a foundation for 6G. Abroad, over 60 companies, from operators and vendors to chipset suppliers and device makers, have signed a GSMA statement on the band’s readiness.

One piece sits beyond any operator’s control. A band is only useful to people whose phones and routers can tune to it. TDRA plans to add Upper 6 GHz, known in standards as n104, to national type-approval rules, with the first devices due from September 2026. How fast those handsets reach shop shelves will decide when the 6G band in the UAE turns from a modelled result into something a customer can measure.

Share this post

on your favourite social platforms

or copy the link