Expect Anthropic to continue refining its blocking mechanisms, likely exploring more sophisticated AI-driven detection systems and stricter enforcement against third-party facilitators. However, the ingenuity of users, coupled with the strong demand for advanced AI, suggests that workarounds will persist and evolve. This situation could push the U.S. government to consider even broader or more aggressive export controls on AI technology, while simultaneously incentivizing China to accelerate its domestic AI development to reduce reliance on foreign models. The cost of maintaining these digital borders will only grow, both for Anthropic and for the broader global AI ecosystem.
Image: courtesy of Wired
The Cat-and-Mouse Game: Chinese Users Keep Bypassing Anthropic's AI Restrictions
Anthropic, a leading AI safety company, is locked in an escalating battle to prevent access to its advanced Claude AI models from China. Despite increasingly strict geolocation and identity verification measures, Chinese users continue to find creative ways to circumvent these blocks, often through virtual private networks (VPNs), proxy services, and the purchase of pre-verified accounts. This ongoing technical arms race unfolds against a backdrop of heightened geopolitical tensions, with the U.S. government imposing export controls on Anthropic's most advanced models, and the company itself claiming to have lost hundreds of millions of dollars in its efforts to comply and restrict access.
Outlook
Background
Anthropic, a company that has positioned itself at the forefront of AI safety, finds itself navigating a complex geopolitical landscape. Its efforts to restrict access to its Claude AI models from China are not merely a technical challenge but a direct response to U.S. government concerns about national security and the potential misuse of advanced AI by authoritarian states. The company has explicitly advocated for export controls, aligning itself with a policy stance that has drawn criticism and irony from Chinese social media users.
Starting in April, Anthropic rolled out identity verification for some Claude users, requiring them to upload government-issued photo IDs through a third-party service called Persona. This was a significant step beyond simple IP-based geolocation. By September 2025, the restrictions extended to organizations, prohibiting access for entities more than 50% owned, directly or indirectly, by companies headquartered in restricted regions like mainland China. This move broadened the scope from individual users to corporate structures.
The U.S. government further escalated the situation on Friday, June 26, 2026, when it issued a directive blocking foreign access to Anthropic’s newest and most advanced AI models, Fable 5 and Mythos 5, even for Anthropic's own foreign-born employees. This demonstrates a deep-seated concern within Washington about the potential security risks associated with these powerful AI systems falling into the wrong hands.
Despite these multi-layered restrictions, Chinese users have consistently found ways to bypass them. Common methods include using VPNs to mask their IP addresses, employing proxy services, or purchasing pre-existing Anthropic accounts from online marketplaces. Reports indicate that Telegram channels have become a hub for facilitating these workarounds, offering users methods to access Claude even after Anthropic's latest crackdowns. Anthropic has stated that these blocking efforts have cost the company hundreds of millions of dollars, highlighting the financial burden of enforcing these geopolitical boundaries.
See also
Precedents
The current struggle between Anthropic and Chinese users mirrors a long-standing history of digital cat-and-mouse games, particularly concerning internet access in China. For decades, Chinese citizens have utilized VPNs and other circumvention tools to bypass the 'Great Firewall,' China's extensive internet censorship system, to access global information and services. This has created a resilient ecosystem of technical workarounds and a user base adept at finding new routes when old ones are blocked.
In the technology sector, the concept of 'technological decoupling' between the U.S. and China has been gaining momentum for several years, particularly in sensitive areas like semiconductors and advanced computing. The export controls on Anthropic's AI models are a direct extension of this trend, moving beyond hardware to restrict access to cutting-edge software and models. Historically, attempts to restrict access to widely desired technologies have often led to the rise of black markets, parallel economies, and an acceleration of domestic alternatives. The current situation with Claude AI access follows this pattern, suggesting that technical controls alone are unlikely to fully stem the flow of information or access without significant, sustained, and evolving effort.
The ongoing struggle to control access to advanced AI models like Claude carries significant implications far beyond Anthropic's balance sheet. For the U.S. government, it represents a critical front in national security, aiming to prevent sophisticated AI from being used by adversaries, particularly those designated as 'Chinese Military Companies.' The ability to effectively enforce these export controls will shape future policy regarding AI development and deployment globally.
For Anthropic, the stakes are existential. The company must balance its stated mission of AI safety and its advocacy for responsible deployment with the commercial reality of a global market. Losing hundreds of millions in revenue to enforcement, while simultaneously being perceived as politically biased by a major market, presents a difficult operational challenge. The company's unique position — seen by some in Washington as 'too woke' yet in Beijing as the 'most hawkish AI company' — underscores the complex political tightrope it walks.
For Chinese users and companies, restricted access to leading-edge AI models could hinder technological progress and innovation, creating a demand vacuum that domestic AI developers will eagerly seek to fill. This could accelerate China's push for AI self-sufficiency, potentially leading to a more bifurcated global AI landscape. Ultimately, this battle highlights the growing tension between the open, collaborative spirit often associated with scientific and technological advancement, and the national security imperatives that increasingly seek to contain and control it.
Scenarios
AnalysisOne possible outcome is that Anthropic and the U.S. government will continue to tighten technical and regulatory controls, leading to an intensified 'arms race' with Chinese users and third-party facilitators. This could involve more advanced AI-driven detection systems, harsher penalties for circumvention, and potentially broader international cooperation on enforcement. However, this approach carries high costs, both financial and in terms of fostering resentment and driving demand for black market solutions.
Another scenario suggests that the persistent circumvention, coupled with the high financial cost to Anthropic, could lead to a re-evaluation of the efficacy and sustainability of current restrictions. If the 'hundreds of millions' in losses continue to mount without a significant reduction in unauthorized access, Anthropic or policymakers might explore alternative strategies. This could involve focusing resources on detecting and mitigating misuse of AI rather than attempting to completely block access, or developing more nuanced, tiered access policies.
A third, more speculative, outcome is that the difficulty in controlling access to advanced AI models like Claude could accelerate the development and adoption of robust domestic AI alternatives within China. Faced with consistent restrictions, Chinese companies and researchers may be further incentivized to invest heavily in developing their own foundational models that are comparable in capability, thereby reducing their reliance on foreign technologies and potentially leading to a more fragmented global AI ecosystem.
Timeline
2025-09-01
Expanded Ownership Restrictions
Anthropic extended its restrictions to prohibit access for organizations more than 50% owned, directly or indirectly, by companies headquartered in unsupported regions, including mainland China.
2026-04-01
Identity Verification Rolled Out
Anthropic introduced identity verification for some Claude users, requiring government-issued photo IDs through a third-party service, Persona, to tighten geolocation controls.
2026-06-26
U.S. Export Controls on Fable 5 and Mythos 5
The U.S. government issued a directive blocking foreign access to Anthropic’s newest AI models, Fable 5 and Mythos 5, due to security risks, impacting even Anthropic's foreign-born employees.
Frequently Asked Questions
Anthropic is restricting access primarily due to U.S. government directives and national security concerns. The U.S. government has expressed worries about advanced AI models being accessed and potentially misused by entities in restricted regions, including those designated as Chinese Military Companies. Anthropic has also publicly advocated for export controls to prevent misuse by authoritarian nations.
Discussion
Be the first to share your thoughts.