Case anthropic-gtg-16005-alibaba-distillModel & data leakUnited States
GTG-16005: Anthropic says Alibaba ran 151M-exchange distillation of Claude to train Qwen
Anthropic attributes to Alibaba's Qwen team a campaign forcing chain-of-thought output via a fixed prompt, peaking near 3M exchanges a day from 3,500+ fake accounts, 151M+ exchanges May to July 2026.
CriticalAttack on AIStatus: OngoingVerified case file
Open in the interactive mapCite this case
- Reported
- Sep 10, 2026
- Location
- United States
- Target
- Anthropic Claude Opus 4.6 and 4.7 reasoning outputs
- Severity
- Critical
- Triggered by
- Alibaba Qwen / Tongyi Lab (per Anthropic)
- Actor type
- Criminal
- Reported loss
- None reported
- CVE
- None
What happened
Anthropic attributes to Alibaba's Qwen team a campaign forcing chain-of-thought output via a fixed prompt, peaking near 3M exchanges a day from 3,500+ fake accounts, 151M+ exchanges May to July 2026.
Impact: Largest distillation campaign Anthropic has measured; transcripts reportedly used to train Qwen models.
Who triggered it
Actor: Alibaba Qwen / Tongyi Lab (per Anthropic)
Attribution: Anthropic attributes with metadata-based proxy-network analysis; Alibaba's response not seen in sources.
Motive: Training data for Qwen 3.5, 3.6 and 3.7 at a fraction of the cost
How it happened
Attack vector: Thousands of fraudulent accounts behind proxy networks forcing reasoning output and harvesting transcripts for fine-tuning
Root cause: API access at scale with weak identity checks on accounts from unsupported regions
Timeline
- Campaign peaks near 3 million exchanges per day
- Over 151 million exchanges observed since May; actor shifts to a second account pool after bans
- Anthropic publishes attribution
Response
Accounts banned, extraction classifiers strengthened, reasoning summarised, identity verification imposed on suspicious accounts.
Current status: Ongoing
Sources (open in a new tab)
- Anthropic September 2026 threat report anthropic.com
- Unite.AI summary unite.ai
Cite this case
GTG-16005: Anthropic says Alibaba ran 151M-exchange distillation of Claude to train Qwen. AI Threat Watch, case anthropic-gtg-16005-alibaba-distill, Sep 10, 2026. https://aithreat.io/cases/anthropic-gtg-16005-alibaba-distill/
Permanent link: https://aithreat.io/cases/anthropic-gtg-16005-alibaba-distill/