Bitcoin Red Team returns to Chinese AI models after OpenAI block
The volunteer initiative Bitcoin Red Team, which audits the security of the Bitcoin ecosystem using artificial intelligence, has encountered an unexpected obstacle. The project's leader, AnchorWatch CEO Rob Hamilton, began integrating OpenAI's cyber capabilities into the workflow on August 8, but by the next morning discovered that access to the tools had been restricted.
This is particularly notable given that Hamilton completed the full KYC procedure and finished onboarding several months ago to legally use the American company's advanced developments. Despite this, the system still blocked his activity, forcing the team to quickly switch to open Chinese models.
Asymmetry in favor of attackers
According to Hamilton, the restrictions prevented completing the audit of code changes and checking for other potential vulnerabilities. He emphasizes that such policies by American AI giants create a dangerous imbalance: defenders are deprived of the most powerful tools, while attackers are not bound by any corporate restrictions.
Interestingly, Anthropic's policy regarding trusted access has not yet caused problems. However, Hamilton ironically noted that he would likely be downgraded from the Fable model to Opus, which is essentially the same problem in a different form.
Scope of the audit and costs
For context, Bitcoin Red Team is a volunteer association of 25 specialists who combine manual review with AI analysis of hundreds of open-source repositories related to Bitcoin. As of August 9, the team had already discovered 1,288 critical and high-level vulnerabilities in the ecosystem of the first cryptocurrency.
The financial side of the issue is also telling. Total AI inference costs have exceeded $58,000, and despite the impressive results of OpenAI and Anthropic's cyber models, about 75% of expenses still go to the Chinese model Kimi K3. The share of Qwen 3.8 is also continuing to grow.
Hamilton himself humorously noted that he and his colleague under the nickname calle are beginning a "very Chinese period of life."
My comment: The situation clearly demonstrates that geopolitical restrictions in the AI field primarily hit the security of open-source software. While American regulators and corporations argue over who can be trusted with advanced models, Chinese counterparts are already becoming the de facto standard for critical tasks such as code auditing. This creates a paradoxical situation where Western restrictions only accelerate China's technological dominance in the niche of specialized AI tools.