AI Theft Bombshell Rattles Alibaba

Alibaba barred employees from using Anthropic’s coding tool after Anthropic told Congress it traced 28.8 million suspicious exchanges to 25,000 fake accounts tied to Alibaba’s network.

Story Highlights

Anthropic alleged a record-scale “distillation” campaign targeting Claude’s coding and reasoning features.
Alibaba denied using proprietary model outputs for training and fought a U.S. military-link label in court.
The campaign allegedly ran 44 days and used commercial proxies to evade regional blocks.
U.S. media and lawmakers framed the clash as theft amid rising talk of sanctions on Chinese firms.

What Triggered Alibaba’s Company-Wide Ban

Anthropic sent a letter to the United States Senate Banking Committee on June 10, 2026, alleging a record “distillation” attack that sought to extract Claude’s agentic reasoning, software engineering, and long-horizon planning abilities. The company said operators linked to Alibaba and its Qwen lab pushed 28.8 million exchanges through about 25,000 fraudulent accounts between April 22 and June 5. Anthropic called it the largest such operation it has seen, nearly double earlier campaigns it tied to other Chinese labs.

Anthropic said the operators used commercial proxies to bypass blocks that restrict access from China-based entities, a tactic the firm also described in prior disclosures about similar schemes. The company highlighted unusual query patterns that focused on specific capabilities, which it argues signal deliberate extraction rather than normal use. These details, if accurate, mark a scale that could speed up a rival model’s skills while sidestepping the high cost of research and evaluation.

Alibaba’s Denial And The Evidence Gap

Alibaba denied using outputs from proprietary models to train its systems and said it follows intellectual property laws. The company, however, did not release network logs or internal records that rebut Anthropic’s counts or proxy claims. Anthropic’s figures have not been verified by an independent auditor, and its letter infers Alibaba affiliation rather than proving employment ties. This leaves a sharp dispute with detailed claims on one side and a broad denial on the other.

Alibaba is also challenging the United States Department of Defense label that describes it as linked to China’s military strategy, arguing the designation lacks factual basis. That fight, now in federal court in San Jose, shapes how lawmakers and the public view the firm, regardless of the technical merits in the distillation case. Media coverage has leaned on terms like “theft,” which raises the political stakes before any neutral review of the data.

Why This Clash Matters For Workers, Builders, And The Law

Developers and businesses now face a two-front risk: losing model know-how to mass querying, and getting locked out by tighter access rules. United States firms, pressed by investors and Congress, are moving to stricter screening, rate limits, and watermarks to stop industrial-scale harvesting. Past incidents tied to labs like DeepSeek, Moonshot, and MiniMax showed similar playbooks with thousands of fake accounts and millions of prompts, but produced little formal enforcement due to cross-border limits.

alibaba is banning claude code from july 10…

employees will be pushed toward qoder instead.

the official reason:

• hidden fingerprinting mechanisms
• alleged backdoor risks
• concerns over telemetry and data collection

this comes days after anthropic accused alibaba of:… https://t.co/3qExUnbWLO pic.twitter.com/X5OSnwmvjb

— Yum⋆₊˚ (@yuhasbeentaken) July 4, 2026

Both right and left share a core worry here: powerful players write the rules after the damage is done. If Anthropic’s claims hold up, the message is clear—unauthorized distillation can drain years of work and tilt the field toward those who can hide behind proxies. If the claims overreach, then policy and sanctions may again move faster than facts, hurting honest users and raising costs across the board. Independent audits of logs and models would help settle the core questions.

What To Watch Next

Lawmakers are weighing sanctions aimed at firms accused of exploiting American models. Regulators may also push stronger know-your-customer checks and third-party reviews of traffic patterns for major labs. A credible audit could trace proxy routes, confirm account clusters, and compare model outputs to test for distilled behavior. Until then, the cycle repeats: accusation, denial, and policy shock that lands hardest on rank-and-file workers trying to build and ship real tools.