Anthropic says Alibaba ran a large-scale campaign to illicitly access Claude through thousands of fraudulent accounts. This is precisely the scenario the Permission Layer was designed to prevent β and it just became the industry’s most cited proof of concept.
Bloomberg reported June 24, 2026 that Anthropic has formally accused Alibaba of orchestrating a “large-scale” effort to illicitly access its Claude AI model. The mechanism: thousands of fraudulent user accounts, systematically created to bypass Anthropic’s access controls and extract capability from a US frontier model.
Anthropic has not disclosed the technical depth of what was accessed β whether it was API outputs, model weights, or training signal extraction. But the scale of the account operation signals this was not opportunistic scraping. It was coordinated, state-adjacent, and systematic.
The timing is pointed. Anthropic announced mandatory passport and national ID verification for all Claude API users starting July 8. The Fable 5 generative media model was shut down to foreign nationals on June 12 under US government instruction. The Permission Layer is no longer theory β it is policy, with teeth.
The Permission Layer Timeline
2025 β Background
Glasswing Ventures documents 200+ organizations blocked from frontier AI access via early permissioning frameworks. The scale of foreign-entity probing becomes visible for the first time.
June 12, 2026
US government orders Fable 5 shut down to foreign nationals β the first explicit nationality-based AI access cut.
June 19, 2026
DeepSeek V4-Pro launches, trained on Huawei Ascend 950 chips. China proves it can produce frontier AI domestically β and may have been supplementing that effort with access to Western models.
June 24, 2026
Bloomberg: Anthropic accuses Alibaba of using thousands of fraudulent accounts to illicitly access Claude at scale.
July 8, 2026
Anthropic mandates passport/national ID verification for all Claude API users. The Permission Layer goes hard.
The Structural Read
The Alibaba accusation is not primarily a corporate dispute. It is evidence that the Permission Layer β the emerging infrastructure of government and corporate controls that determines who can access frontier AI β is now the most contested real estate in tech.
The strategic logic of what Alibaba allegedly did is straightforward: China’s own models, however impressive, are being trained in partial isolation from the most capable Western systems. Access to Claude β even through back channels β provides alignment signal, capability benchmarks, and potential distillation targets. It is intelligence gathering by another name.
What makes this story structurally significant is the “belt and suspenders” reality it exposes. China committed $295 billion to domestic AI infrastructure. It launched DeepSeek V4-Pro on Huawei chips. It is building its own sovereign stack, end to end. And simultaneously, according to Anthropic, it was trying to access Claude through the back door. The two strategies are not contradictory β they are complementary. Build your own; steal the rest.
The key insight: Export controls and chip restrictions slow the hardware side of China’s AI stack. But without identity-verified access controls on the model side, frontier AI capability is still leaking β through APIs, through fraudulent accounts, through the front door left open by frictionless sign-up.
Anthropic / Bloomberg
“Alibaba waged a large-scale effort to illicitly access Claude AI, using thousands of fraudulent accounts.”
Three Implications
IDENTITY IS NOW THE MOAT
Anthropic’s July 8 ID verification mandate is not compliance theater. It is the minimum viable fence around frontier AI. Any lab that does not implement equivalent controls is effectively running an open API for state-adjacent actors. The industry will follow β or face the same accusations.
DEEPSEEK + CLAUDE = BELT AND SUSPENDERS
China’s official strategy β the $295B infrastructure plan, Huawei Ascend, DeepSeek V4-Pro β is the public layer. Alleged Claude access via fake accounts is the covert layer. State-level AI strategy does not rely on a single approach. The export-control regime was designed to stop one; it was not designed to stop both simultaneously.
THE PERMISSION LAYER IS BECOMING GEOPOLITICAL INFRASTRUCTURE
Glasswing’s 200-organization block count, Fable 5’s forced shutdown, and now the Alibaba accusation are not separate stories. They are data points in the same thesis: the access layer for frontier AI has become as strategically contested as physical chip manufacturing. The companies that control it β and the governments that regulate it β are writing the rules of the next decade.
The Permission Layer β Who Holds What
US Government (Fable 5 shutdown)
ESCALATING
First nationality-based model access cut on record. Sets precedent for future model restrictions by executive order.
Anthropic (ID verification mandate)
PROACTIVE
Passport/ID verification from July 8. First major US lab to implement real-identity binding at the API layer.
Glasswing Ventures (200+ org blocks)
DOCUMENTED
Early permissioning infrastructure had already identified and blocked over 200 organizations from frontier model access β before this incident surfaced publicly.
Building sovereign AI infrastructure in public. Allegedly accessing US frontier models through the back door in private. Both tracks running in parallel.
Business Engineer Framework
The Permission Layer β Who Controls AI Access Controls the Future
The Alibaba incident is the Permission Layer thesis made concrete. Governments, labs, and enterprise security teams are all converging on the same realization: who can access frontier AI is now a geopolitical question, not just a product decision. The Harness Society unpacks the full architecture of how access control becomes competitive moat.
China’s AI strategy was never single-threaded. While the world watched the $295 billion infrastructure plan and the Huawei chip story, Alibaba was allegedly running thousands of fake accounts to extract capability from Claude. The belt (DeepSeek on domestic chips) and the suspenders (fraudulent Claude access) were operating simultaneously. Anthropic’s July 8 identity verification mandate is not a bureaucratic inconvenience β it is the minimum viable response to a threat that was already active. The Permission Layer just found its most important case study.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
