Bitcoin Red Team is forced to return to Chinese AI models due to OpenAI sanctions.
The volunteer initiative Bitcoin Red Team, specializing in finding vulnerabilities in the open-source Bitcoin ecosystem, has encountered a serious barrier from American AI providers. The project lead, AnchorWatch CEO Rob Hamilton, reported that on August 8 he successfully integrated OpenAI's cyber models into the workflow, but access was blocked the very next morning, despite passing KYC procedures and completing full onboarding.
This incident forced the team to quickly switch to open Chinese models, which, according to Hamilton, are becoming a critically important tool for defenders. He highlights the paradox of the situation: American AI companies, by imposing restrictions, deprive legitimate security researchers of access to the most powerful algorithms, while malicious actors are not bound by any regulatory frameworks and continue to use advanced technologies to their full extent.
When asked about Anthropic's policy regarding trusted access, Hamilton noted that no problems have arisen so far, but ironically added: "I'll probably be downgraded from the Fable model to Opus, so in a sense it's the same thing." This statement underscores a systemic problem: even with loyal treatment from providers, the hierarchy of access to models can change unpredictably, creating instability for research projects.
Let me remind you that Bitcoin Red Team is a volunteer group of 25 experts that combines manual code review with AI analysis. As of August 9, they had managed to identify 1,288 critical and high-level vulnerabilities in Bitcoin repositories. Hamilton and his colleague under the handle calle have already joked that they are entering a "very Chinese period of life."
Notably, despite the blocks, the team continues to use Western cyber models when possible, noting their "stunning results." However, the lion's share of AI expenses—almost 75%—still goes to the Chinese model Kimi K3, and the share of Qwen 3.8 from Alibaba is also steadily growing. The project's total AI costs have already exceeded $58,000, which speaks to the scale and seriousness of the work being done.
My analysis: This situation is a vivid example of how geopolitical pressure on the technological front backfires on those who initiated the restrictions. By trying to contain China, American companies are effectively pushing key security researchers into the arms of competitors, stimulating the development and adoption of Chinese AI solutions in critically important infrastructure. This not only undermines U.S. leadership in AI but also creates long-term risks for global cybersecurity.